201.
Otten M; Jagesar A R; Dam T A; Biesheuvel L A; den Hengst F; Ziesemer K A; Thoral P J; de Grooth H; Girbes A R; François-Lavet V; others
Does reinforcement learning improve outcomes for critically ill patients? A systematic review and level-of-readiness assessment Journal Article
In: Critical Care Medicine, vol. 52, no. 2, pp. e79–e88, 2024.
@article{otten2024does,
title = {Does reinforcement learning improve outcomes for critically ill patients? A systematic review and level-of-readiness assessment},
author = {Otten, Martijn and Jagesar, Ameet R and Dam, Tariq A and Biesheuvel, Laurens A and den Hengst, Floris and Ziesemer, Kirsten A and Thoral, Patrick J and de Grooth, Harm-Jan and Girbes, Armand RJ and Fran{ç}ois-Lavet, Vincent and others},
year = {2024},
date = {2024-01-01},
journal = {Critical Care Medicine},
volume = {52},
number = {2},
pages = {e79–e88},
publisher = {LWW},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
200.
Hengst F; Wolter R; Altmeyer P; Kaygan A
Conformal Intent Classification and Clarification for Fast and Accurate Intent Recognition Proceedings Article
In: Findings of the Association for Computational Linguistics: NAACL 2024, pp. 2412–2432, 2024.
@inproceedings{hengst2024conformal,
title = {Conformal Intent Classification and Clarification for Fast and Accurate Intent Recognition},
author = {Hengst, Floris and Wolter, Ralf and Altmeyer, Patrick and Kaygan, Arda},
year = {2024},
date = {2024-01-01},
booktitle = {Findings of the Association for Computational Linguistics: NAACL 2024},
pages = {2412–2432},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
199.
Chen P; Baez Santamaria S; de Boer M H; den Hengst F; Kamphorst B A; Smit Q; Wang S; Wolff J
Intelligent Support Systems for Lifestyle Change: Integrating Dialogue, Information Extraction, and Reasoning Book Section
In: HHAI 2024: Hybrid Human AI Systems for the Social Good, pp. 457–459, IOS Press, 2024.
@incollection{chen2024intelligent,
title = {Intelligent Support Systems for Lifestyle Change: Integrating Dialogue, Information Extraction, and Reasoning},
author = {Chen, Pei-Yu and Baez Santamaria, Selene and de Boer, Maaike HT and den Hengst, Floris and Kamphorst, Bart A and Smit, Quirine and Wang, Shihan and Wolff, Johanna},
year = {2024},
date = {2024-01-01},
booktitle = {HHAI 2024: Hybrid Human AI Systems for the Social Good},
pages = {457–459},
publisher = {IOS Press},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
198.
Zhang J E; Hilpert B; Broekens J; Jokinen J P
Simulating Emotions With an Integrated Computational Model of Appraisal and Reinforcement Learning Proceedings Article
In: Proceedings of the CHI Conference on Human Factors in Computing Systems, pp. 1–12, 2024.
@inproceedings{zhang2024simulating,
title = {Simulating Emotions With an Integrated Computational Model of Appraisal and
Reinforcement Learning},
author = {Zhang, Jiayi Eurus and Hilpert, Bernhard and Broekens, Joost and Jokinen, Jussi
PP},
year = {2024},
date = {2024-01-01},
booktitle = {Proceedings of the CHI Conference on Human Factors in Computing Systems},
pages = {1–12},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
197.
Meulen R; Verbrugge R; Duijn M
Common ground provides a mental shortcut in agent-agent interaction Journal Article
In: pp. 281-290, 2024.
@article{meulen2024common,
title = {Common ground provides a mental shortcut in agent-agent interaction},
author = {Meulen, Ramira van der and Verbrugge, Rineke and Duijn, Max van},
year = {2024},
date = {2024-01-01},
booktitle = {HHAI 2024: Hybrid Human AI Systems for the Social Good},
pages = {281-290},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
196.
Van Leeuwen L; Verbrugge R; Verheij B; Renooij S
Building a Stronger Case: Combining Evidence and Law in Scenario-Based Bayesian Networks Proceedings Article
In: 3rd International Conference on Hybrid Human-Artificial Intelligence, HHAI 2024, pp. 291–299, IOS Press 2024.
@inproceedings{van2024building,
title = {Building a Stronger Case: Combining Evidence and Law in Scenario-Based Bayesian
Networks},
author = {Van Leeuwen, Ludi and Verbrugge, Rineke and Verheij, Bart and Renooij, Silja},
year = {2024},
date = {2024-01-01},
booktitle = {3rd International Conference on Hybrid Human-Artificial Intelligence, HHAI
2024},
pages = {291–299},
organization = {IOS Press},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
195.
Nguyen T; Chatterjee S; MacAvaney S; Mackie I; Dalton J; Yates A
DyVo: Dynamic Vocabularies for Learned Sparse Retrieval with Entities Proceedings Article
In: Al-Onaizan, Yaser; Bansal, Mohit; Chen, Yun-Nung (Ed.): Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, EMNLP 2024, Miami, FL, USA, November 12-16, 2024, pp. 767–783, Association for Computational Linguistics, 2024.
@inproceedings{DBLP:conf/emnlp/NguyenCMM0Y24,
title = {DyVo: Dynamic Vocabularies for Learned Sparse Retrieval with Entities},
author = {Thong Nguyen and
Shubham Chatterjee and
Sean MacAvaney and
Iain Mackie and
Jeff Dalton and
Andrew Yates},
editor = {Yaser Al{-}Onaizan and
Mohit Bansal and
Yun{-}Nung Chen},
url = {https://aclanthology.org/2024.emnlp-main.45},
year = {2024},
date = {2024-01-01},
booktitle = {Proceedings of the 2024 Conference on Empirical Methods in Natural
Language Processing, EMNLP 2024, Miami, FL, USA, November 12-16,
2024},
pages = {767–783},
publisher = {Association for Computational Linguistics},
abstract = {Learned Sparse Retrieval (LSR) models use vocabularies from pre-trained transformers, which often split entities into nonsensical fragments. Splitting entities diminishes retrieval accuracy and limits the model’s ability to incorporate up-to-date world knowledge not included in the training data. In this work, we enhance the LSR vocabulary with Wikipedia concepts and entities, enabling the model to resolve ambiguities more effectively and stay current with evolving knowledge. Central to our approach is a Dynamic Vocabulary (DyVo) head, which leverages existing entity embeddings and an entity retrieval component that identifies entities relevant to a query or document. We use the DyVo head to generate entity weights, which are then merged with word piece weights to create joint representations for efficient indexing and retrieval using an inverted index. In experiments across three entity-rich document ranking datasets, the resulting DyVo model substantially outperforms several state-of-the-art baselines.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Learned Sparse Retrieval (LSR) models use vocabularies from pre-trained transformers, which often split entities into nonsensical fragments. Splitting entities diminishes retrieval accuracy and limits the model’s ability to incorporate up-to-date world knowledge not included in the training data. In this work, we enhance the LSR vocabulary with Wikipedia concepts and entities, enabling the model to resolve ambiguities more effectively and stay current with evolving knowledge. Central to our approach is a Dynamic Vocabulary (DyVo) head, which leverages existing entity embeddings and an entity retrieval component that identifies entities relevant to a query or document. We use the DyVo head to generate entity weights, which are then merged with word piece weights to create joint representations for efficient indexing and retrieval using an inverted index. In experiments across three entity-rich document ranking datasets, the resulting DyVo model substantially outperforms several state-of-the-art baselines.
194.
Wu D; Lei Y; Yates A; Monz C
Representational Isomorphism and Alignment of Multilingual Large Language Models Proceedings Article
In: Al-Onaizan, Yaser; Bansal, Mohit; Chen, Yun-Nung (Ed.): Findings of the Association for Computational Linguistics: EMNLP 2024, Miami, Florida, USA, November 12-16, 2024, pp. 14074–14085, Association for Computational Linguistics, 2024.
@inproceedings{DBLP:conf/emnlp/WuLYM24,
title = {Representational Isomorphism and Alignment of Multilingual Large Language
Models},
author = {Di Wu and
Yibin Lei and
Andrew Yates and
Christof Monz},
editor = {Yaser Al{-}Onaizan and
Mohit Bansal and
Yun{-}Nung Chen},
url = {https://aclanthology.org/2024.findings-emnlp.823},
year = {2024},
date = {2024-01-01},
booktitle = {Findings of the Association for Computational Linguistics: EMNLP
2024, Miami, Florida, USA, November 12-16, 2024},
pages = {14074–14085},
publisher = {Association for Computational Linguistics},
abstract = {In this paper, we investigate the capability of Large Language Models (LLMs) to represent texts in multilingual contexts. Our findings show that sentence representations derived from LLMs exhibit a high degree of isomorphism across languages.This existing isomorphism can facilitate representational alignments in zero-shot and few-shot settings.Specifically, by applying a contrastive objective at the representation level with only a small number of translation pairs (e.g., 100), we substantially improve models’ performance on Semantic Textual Similarity (STS) tasks across languages. This representation-level approach proves to be more efficient and effective for semantic alignment than continued pretraining or instruction tuning. Interestingly, we also observe substantial STS improvements within individual languages, even without a monolingual objective specifically designed for this purpose.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
In this paper, we investigate the capability of Large Language Models (LLMs) to represent texts in multilingual contexts. Our findings show that sentence representations derived from LLMs exhibit a high degree of isomorphism across languages.This existing isomorphism can facilitate representational alignments in zero-shot and few-shot settings.Specifically, by applying a contrastive objective at the representation level with only a small number of translation pairs (e.g., 100), we substantially improve models’ performance on Semantic Textual Similarity (STS) tasks across languages. This representation-level approach proves to be more efficient and effective for semantic alignment than continued pretraining or instruction tuning. Interestingly, we also observe substantial STS improvements within individual languages, even without a monolingual objective specifically designed for this purpose.
193.
Bleeker M J R; Hendriksen M; Yates A; de Rijke M
Demonstrating and Reducing Shortcuts in Vision-Language Representation Learning Journal Article
In: CoRR, vol. abs/2402.17510, 2024.
@article{DBLP:journals/corr/abs-2402-17510,
title = {Demonstrating and Reducing Shortcuts in Vision-Language Representation
Learning},
author = {Maurits J. R. Bleeker and
Mariya Hendriksen and
Andrew Yates and
Maarten de Rijke},
url = {https://doi.org/10.48550/arXiv.2402.17510},
doi = {10.48550/ARXIV.2402.17510},
year = {2024},
date = {2024-01-01},
journal = {CoRR},
volume = {abs/2402.17510},
abstract = {Vision-language models (VLMs) mainly rely on contrastive training to learn general-purpose representations of images and captions. We focus on the situation when one image is associated with several captions, each caption containing both information shared among all captions and unique information per caption about the scene depicted in the image. In such cases, it is unclear whether contrastive losses are sufficient for learning task-optimal representations that contain all the information provided by the captions or whether the contrastive learning setup encourages the learning of a simple shortcut that minimizes contrastive loss. We introduce synthetic shortcuts for vision-language: a training and evaluation framework where we inject synthetic shortcuts into image-text data. We show that contrastive VLMs trained from scratch or fine-tuned with data containing these synthetic shortcuts mainly learn features that represent the shortcut. Hence, contrastive losses are not sufficient to learn task-optimal representations, i.e., representations that contain all task-relevant information shared between the image and associated captions. We examine two methods to reduce shortcut learning in our training and evaluation framework: (i) latent target decoding and (ii) implicit feature modification. We show empirically that both methods improve performance on the evaluation task, but only partly reduce shortcut learning when training and evaluating with our shortcut learning framework. Hence, we show the difficulty and challenge of our shortcut learning framework for contrastive vision-language representation learning.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Vision-language models (VLMs) mainly rely on contrastive training to learn general-purpose representations of images and captions. We focus on the situation when one image is associated with several captions, each caption containing both information shared among all captions and unique information per caption about the scene depicted in the image. In such cases, it is unclear whether contrastive losses are sufficient for learning task-optimal representations that contain all the information provided by the captions or whether the contrastive learning setup encourages the learning of a simple shortcut that minimizes contrastive loss. We introduce synthetic shortcuts for vision-language: a training and evaluation framework where we inject synthetic shortcuts into image-text data. We show that contrastive VLMs trained from scratch or fine-tuned with data containing these synthetic shortcuts mainly learn features that represent the shortcut. Hence, contrastive losses are not sufficient to learn task-optimal representations, i.e., representations that contain all task-relevant information shared between the image and associated captions. We examine two methods to reduce shortcut learning in our training and evaluation framework: (i) latent target decoding and (ii) implicit feature modification. We show empirically that both methods improve performance on the evaluation task, but only partly reduce shortcut learning when training and evaluating with our shortcut learning framework. Hence, we show the difficulty and challenge of our shortcut learning framework for contrastive vision-language representation learning.
192.
Nguyen T; Hendriksen M; Yates A; de Rijke M
Multimodal Learned Sparse Retrieval with Probabilistic Expansion Control Proceedings Article
In: Goharian, Nazli; Tonellotto, Nicola; He, Yulan; Lipani, Aldo; McDonald, Graham; Macdonald, Craig; Ounis, Iadh (Ed.): Advances in Information Retrieval - 46th European Conference on Information Retrieval, ECIR 2024, Glasgow, UK, March 24-28, 2024, Proceedings, Part II, pp. 448–464, Springer, 2024.
@inproceedings{DBLP:conf/ecir/NguyenHYR24,
title = {Multimodal Learned Sparse Retrieval with Probabilistic Expansion Control},
author = {Thong Nguyen and
Mariya Hendriksen and
Andrew Yates and
Maarten de Rijke},
editor = {Nazli Goharian and
Nicola Tonellotto and
Yulan He and
Aldo Lipani and
Graham McDonald and
Craig Macdonald and
Iadh Ounis},
url = {https://doi.org/10.1007/978-3-031-56060-6_29},
doi = {10.1007/978-3-031-56060-6_29},
year = {2024},
date = {2024-01-01},
booktitle = {Advances in Information Retrieval - 46th European Conference on Information
Retrieval, ECIR 2024, Glasgow, UK, March 24-28, 2024, Proceedings,
Part II},
volume = {14609},
pages = {448–464},
publisher = {Springer},
series = {Lecture Notes in Computer Science},
abstract = {Learned sparse retrieval (LSR) is a family of neural methods that encode queries and documents into sparse lexical vectors that can be indexed and retrieved efficiently with an inverted index. We explore the application of LSR to the multi-modal domain, with a focus on text-image retrieval. While LSR has seen success in text retrieval, its application in multimodal retrieval remains underexplored. Current approaches like LexLIP and STAIR require complex multi-step training on massive datasets. Our proposed approach efficiently transforms dense vectors from a frozen dense model into sparse lexical vectors. We address issues of high dimension co-activation and semantic deviation through a new training algorithm, using Bernoulli random variables to control query expansion. Experiments with two dense models (BLIP, ALBEF) and two datasets (MSCOCO, Flickr30k) show that our proposed algorithm effectively reduces co-activation and semantic deviation. Our best-performing sparsified model outperforms state-of-the-art text-image LSR models with a shorter training time and lower GPU memory requirements. Our approach offers an effective solution for training LSR retrieval models in multimodal settings. Our code and model checkpoints are available at github.com/thongnt99/lsr-multimodal.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Learned sparse retrieval (LSR) is a family of neural methods that encode queries and documents into sparse lexical vectors that can be indexed and retrieved efficiently with an inverted index. We explore the application of LSR to the multi-modal domain, with a focus on text-image retrieval. While LSR has seen success in text retrieval, its application in multimodal retrieval remains underexplored. Current approaches like LexLIP and STAIR require complex multi-step training on massive datasets. Our proposed approach efficiently transforms dense vectors from a frozen dense model into sparse lexical vectors. We address issues of high dimension co-activation and semantic deviation through a new training algorithm, using Bernoulli random variables to control query expansion. Experiments with two dense models (BLIP, ALBEF) and two datasets (MSCOCO, Flickr30k) show that our proposed algorithm effectively reduces co-activation and semantic deviation. Our best-performing sparsified model outperforms state-of-the-art text-image LSR models with a shorter training time and lower GPU memory requirements. Our approach offers an effective solution for training LSR retrieval models in multimodal settings. Our code and model checkpoints are available at github.com/thongnt99/lsr-multimodal.
191.
Lei Y; Cao Y; Zhou T; Shen T; Yates A
Corpus-Steered Query Expansion with Large Language Models Proceedings Article
In: Graham, Yvette; Purver, Matthew (Ed.): Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2024 - Volume 2: Short Papers, St. Julian's, Malta, March 17-22, 2024, pp. 393–401, Association for Computational Linguistics, 2024.
@inproceedings{DBLP:conf/eacl/LeiCZSY24,
title = {Corpus-Steered Query Expansion with Large Language Models},
author = {Yibin Lei and
Yu Cao and
Tianyi Zhou and
Tao Shen and
Andrew Yates},
editor = {Yvette Graham and
Matthew Purver},
url = {https://aclanthology.org/2024.eacl-short.34},
year = {2024},
date = {2024-01-01},
booktitle = {Proceedings of the 18th Conference of the European Chapter of the
Association for Computational Linguistics, EACL 2024 - Volume 2:
Short Papers, St. Julian's, Malta, March 17-22, 2024},
pages = {393–401},
publisher = {Association for Computational Linguistics},
abstract = {Recent studies demonstrate that query expansions generated by large language models (LLMs) can considerably enhance information retrieval systems by generating hypothetical documents that answer the queries as expansions. However, challenges arise from misalignments between the expansions and the retrieval corpus, resulting in issues like hallucinations and outdated information due to the limited intrinsic knowledge of LLMs. Inspired by Pseudo Relevance Feedback (PRF), we introduce Corpus-Steered Query Expansion (CSQE) to promote the incorporation of knowledge embedded within the corpus. CSQE utilizes the relevance assessing capability of LLMs to systematically identify pivotal sentences in the initially-retrieved documents. These corpus-originated texts are subsequently used to expand the query together with LLM-knowledge empowered expansions, improving the relevance prediction between the query and the target documents. Extensive experiments reveal that CSQE exhibits strong performance without necessitating any training, especially with queries for which LLMs lack knowledge.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Recent studies demonstrate that query expansions generated by large language models (LLMs) can considerably enhance information retrieval systems by generating hypothetical documents that answer the queries as expansions. However, challenges arise from misalignments between the expansions and the retrieval corpus, resulting in issues like hallucinations and outdated information due to the limited intrinsic knowledge of LLMs. Inspired by Pseudo Relevance Feedback (PRF), we introduce Corpus-Steered Query Expansion (CSQE) to promote the incorporation of knowledge embedded within the corpus. CSQE utilizes the relevance assessing capability of LLMs to systematically identify pivotal sentences in the initially-retrieved documents. These corpus-originated texts are subsequently used to expand the query together with LLM-knowledge empowered expansions, improving the relevance prediction between the query and the target documents. Extensive experiments reveal that CSQE exhibits strong performance without necessitating any training, especially with queries for which LLMs lack knowledge.
190.
Lei Y; Wu D; Zhou T; Shen T; Cao Y; Tao C; Yates A
Meta-Task Prompting Elicits Embeddings from Large Language Models Proceedings Article
In: Ku, Lun-Wei; Martins, Andre; Srikumar, Vivek (Ed.): Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2024, Bangkok, Thailand, August 11-16, 2024, pp. 10141–10157, Association for Computational Linguistics, 2024.
@inproceedings{DBLP:conf/acl/LeiW00CTY24,
title = {Meta-Task Prompting Elicits Embeddings from Large Language Models},
author = {Yibin Lei and
Di Wu and
Tianyi Zhou and
Tao Shen and
Yu Cao and
Chongyang Tao and
Andrew Yates},
editor = {Lun{-}Wei Ku and
Andre Martins and
Vivek Srikumar},
url = {https://doi.org/10.18653/v1/2024.acl-long.546},
doi = {10.18653/V1/2024.ACL-LONG.546},
year = {2024},
date = {2024-01-01},
booktitle = {Proceedings of the 62nd Annual Meeting of the Association for Computational
Linguistics (Volume 1: Long Papers), ACL 2024, Bangkok, Thailand,
August 11-16, 2024},
pages = {10141–10157},
publisher = {Association for Computational Linguistics},
abstract = {We introduce a new unsupervised text embedding method, Meta-Task Prompting with Explicit One-Word Limitation (MetaEOL), for generating high-quality sentence embeddings from Large Language Models (LLMs) without the need for model fine-tuning. Leveraging meta-task prompting, MetaEOL guides LLMs to produce embeddings through a series of carefully designed prompts that address multiple representational aspects. Our comprehensive experiments demonstrate that embeddings averaged from various meta-tasks are versatile embeddings that yield competitive performance on Semantic Textual Similarity (STS) benchmarks and excel in downstream tasks, surpassing contrastive-trained models. Our findings suggest a new scaling law, offering a versatile and resource-efficient approach for embedding generation across diverse scenarios.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
We introduce a new unsupervised text embedding method, Meta-Task Prompting with Explicit One-Word Limitation (MetaEOL), for generating high-quality sentence embeddings from Large Language Models (LLMs) without the need for model fine-tuning. Leveraging meta-task prompting, MetaEOL guides LLMs to produce embeddings through a series of carefully designed prompts that address multiple representational aspects. Our comprehensive experiments demonstrate that embeddings averaged from various meta-tasks are versatile embeddings that yield competitive performance on Semantic Textual Similarity (STS) benchmarks and excel in downstream tasks, surpassing contrastive-trained models. Our findings suggest a new scaling law, offering a versatile and resource-efficient approach for embedding generation across diverse scenarios.
189.
Krasakis A M; Yates A; Kanoulas E
Contextualizing and Expanding Conversational Queries without Supervision Journal Article
In: ACM Trans. Inf. Syst., vol. 42, no. 3, pp. 77:1–77:30, 2024.
@article{DBLP:journals/tois/KrasakisYK24,
title = {Contextualizing and Expanding Conversational Queries without Supervision},
author = {Antonios Minas Krasakis and
Andrew Yates and
Evangelos Kanoulas},
url = {https://doi.org/10.1145/3632622},
doi = {10.1145/3632622},
year = {2024},
date = {2024-01-01},
journal = {ACM Trans. Inf. Syst.},
volume = {42},
number = {3},
pages = {77:1–77:30},
abstract = {Most conversational passage retrieval systems try to resolve conversational dependencies by using an intermediate query resolution step. To do so, they synthesize conversational data or assume the availability of large-scale question rewriting datasets. To relax those conditions, we propose a zero-shot unified resolution–retrieval approach, that (i) contextualizes and (ii) expands query embeddings using the conversation history and without fine-tuning on conversational data. Contextualization biases the last user question embeddings towards the conversation. Query expansion is used in two ways: (i) abstractive expansion generates embeddings based on the current question and previous history, whereas (ii) extractive expansion tries to identify history term embeddings based on attention weights from the retriever. Our experiments demonstrate the effectiveness of both contextualization and unified expansion in improving conversational retrieval. Contextualization does so mostly by resolving anaphoras to the conversation and bringing their embeddings closer to the important resolution terms that were omitted. By adding embeddings to the query, expansion targets phenomena of ellipsis more explicitly, with our analysis verifying its effectiveness on identifying and adding important resolutions to the query. By combining contextualization and expansion, we find that our zero-shot unified resolution–retrieval methods are competitive and can even outperform supervised methods.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Most conversational passage retrieval systems try to resolve conversational dependencies by using an intermediate query resolution step. To do so, they synthesize conversational data or assume the availability of large-scale question rewriting datasets. To relax those conditions, we propose a zero-shot unified resolution–retrieval approach, that (i) contextualizes and (ii) expands query embeddings using the conversation history and without fine-tuning on conversational data. Contextualization biases the last user question embeddings towards the conversation. Query expansion is used in two ways: (i) abstractive expansion generates embeddings based on the current question and previous history, whereas (ii) extractive expansion tries to identify history term embeddings based on attention weights from the retriever. Our experiments demonstrate the effectiveness of both contextualization and unified expansion in improving conversational retrieval. Contextualization does so mostly by resolving anaphoras to the conversation and bringing their embeddings closer to the important resolution terms that were omitted. By adding embeddings to the query, expansion targets phenomena of ellipsis more explicitly, with our analysis verifying its effectiveness on identifying and adding important resolutions to the query. By combining contextualization and expansion, we find that our zero-shot unified resolution–retrieval methods are competitive and can even outperform supervised methods.
188.
Hou M; Hindriks K V; Eiben G; Baraka K
"Give Me an Example Like This": Episodic Active Reinforcement Learning from Demonstrations Proceedings Article
In: Ahmad, Muneeb Imtiaz; Lohan, Katrin S.; Foster, Mary Ellen; Holthaus, Patrick; Nagai, Yukie (Ed.): Proceedings of the 12th International Conference on Human-Agent Interaction, HAI 2024, Swansea, United Kingdom, November 24-27, 2024, pp. 287–295, ACM, 2024.
@inproceedings{DBLP:conf/hai/HouHEB24,
title = {"Give Me an Example Like This": Episodic Active Reinforcement Learning
from Demonstrations},
author = {Muhan Hou and
Koen V. Hindriks and
Guszti Eiben and
Kim Baraka},
editor = {Muneeb Imtiaz Ahmad and
Katrin S. Lohan and
Mary Ellen Foster and
Patrick Holthaus and
Yukie Nagai},
url = {https://doi.org/10.1145/3687272.3688298},
doi = {10.1145/3687272.3688298},
year = {2024},
date = {2024-01-01},
booktitle = {Proceedings of the 12th International Conference on Human-Agent Interaction,
HAI 2024, Swansea, United Kingdom, November 24-27, 2024},
pages = {287–295},
publisher = {ACM},
abstract = {Reinforcement Learning (RL) has achieved great success in sequential decision-making problems but often requires extensive agent-environment interactions. To improve sample efficiency, methods like Reinforcement Learning from Expert Demonstrations (RLED) incorporate external expert demonstrations to aid agent exploration during the learning process. However, these demonstrations, typically collected from human users, are costly and thus often limited in quantity. Therefore, how to select the optimal set of human demonstrations that most effectively aids learning becomes a critical concern. This paper introduces EARLY (Episodic Active Learning from demonstration querY), an algorithm designed to enable a learning agent to generate optimized queries for expert demonstrations in a trajectory-based feature space. EARLY employs a trajectory-level estimate of uncertainty in the agent’s current policy to determine the optimal timing and content for feature-based queries. By querying episodic demonstrations instead of isolated state-action pairs, EARLY enhances the human teaching experience and achieves better learning performance. We validate the effectiveness of our method across three simulated navigation tasks of increasing difficulty. Results indicate that our method achieves expert-level performance in all three tasks, converging over 50% faster than other four baseline methods when demonstrations are generated by simulated oracle policies. A follow-up pilot user study (N = 18) further supports that our method maintains significantly better convergence with human expert demonstrators, while also providing a better user experience in terms of perceived task load and requiring significantly less human time.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Reinforcement Learning (RL) has achieved great success in sequential decision-making problems but often requires extensive agent-environment interactions. To improve sample efficiency, methods like Reinforcement Learning from Expert Demonstrations (RLED) incorporate external expert demonstrations to aid agent exploration during the learning process. However, these demonstrations, typically collected from human users, are costly and thus often limited in quantity. Therefore, how to select the optimal set of human demonstrations that most effectively aids learning becomes a critical concern. This paper introduces EARLY (Episodic Active Learning from demonstration querY), an algorithm designed to enable a learning agent to generate optimized queries for expert demonstrations in a trajectory-based feature space. EARLY employs a trajectory-level estimate of uncertainty in the agent’s current policy to determine the optimal timing and content for feature-based queries. By querying episodic demonstrations instead of isolated state-action pairs, EARLY enhances the human teaching experience and achieves better learning performance. We validate the effectiveness of our method across three simulated navigation tasks of increasing difficulty. Results indicate that our method achieves expert-level performance in all three tasks, converging over 50% faster than other four baseline methods when demonstrations are generated by simulated oracle policies. A follow-up pilot user study (N = 18) further supports that our method maintains significantly better convergence with human expert demonstrators, while also providing a better user experience in terms of perceived task load and requiring significantly less human time.
187.
Renoux J; Grosinge J; Romeo M; Sabu K M; Baraka K; Kaptelinin V
Communication in Human-AI Interaction (Preface) Proceedings Article
In: Ericson, Petter; Khairova, Nina; Vos, Marina De (Ed.): Proceedings of the Workshops at the Third International Conference on Hybrid Human-Artificial Intelligence co-located with (HHAI 2024), Malmö, Sweden, June 10-11, 2024, pp. 78–83, CEUR-WS.org, 2024.
@inproceedings{DBLP:conf/hhai/RenouxGRSBK24,
title = {Communication in Human-AI Interaction (Preface)},
author = {Jennifer Renoux and
Jasmin Grosinge and
Marta Romeo and
Kiran M. Sabu and
Kim Baraka and
Victor Kaptelinin},
editor = {Petter Ericson and
Nina Khairova and
Marina De Vos},
url = {https://ceur-ws.org/Vol-3825/prefaceW3.pdf},
year = {2024},
date = {2024-01-01},
booktitle = {Proceedings of the Workshops at the Third International Conference
on Hybrid Human-Artificial Intelligence co-located with (HHAI 2024),
Malmö, Sweden, June 10-11, 2024},
volume = {3825},
pages = {78–83},
publisher = {CEUR-WS.org},
series = {CEUR Workshop Proceedings},
abstract = {As Artificially Intelligent systems are becoming more and more present in our surroundings, our ways
of interacting with them are also changing. From commercial chatbots to home assistants and robot
companions, machines are progressively taking up the role of “communicators”, provided with their
own agency, and able to interact with their human counterparts in new ways. This workshop aimed
at gathering experts in fields relevant to the study of AI systems as communicators, including but not
limited to Human-Computer Interaction, Artificial Intelligence, Human-Robot and Human-AI Interaction.
It was organized in order to discuss new challenges brought by this recent shift, compare methods and
perspectives between different fields, and foster long-term collaborations.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
As Artificially Intelligent systems are becoming more and more present in our surroundings, our ways
of interacting with them are also changing. From commercial chatbots to home assistants and robot
companions, machines are progressively taking up the role of “communicators”, provided with their
own agency, and able to interact with their human counterparts in new ways. This workshop aimed
at gathering experts in fields relevant to the study of AI systems as communicators, including but not
limited to Human-Computer Interaction, Artificial Intelligence, Human-Robot and Human-AI Interaction.
It was organized in order to discuss new challenges brought by this recent shift, compare methods and
perspectives between different fields, and foster long-term collaborations.
of interacting with them are also changing. From commercial chatbots to home assistants and robot
companions, machines are progressively taking up the role of “communicators”, provided with their
own agency, and able to interact with their human counterparts in new ways. This workshop aimed
at gathering experts in fields relevant to the study of AI systems as communicators, including but not
limited to Human-Computer Interaction, Artificial Intelligence, Human-Robot and Human-AI Interaction.
It was organized in order to discuss new challenges brought by this recent shift, compare methods and
perspectives between different fields, and foster long-term collaborations.
186.
Christofi K; Baraka K
Uncovering Patterns in Humans that Teach Robots through Demonstrations and Feedback Proceedings Article
In: Grollman, Dan; Broadbent, Elizabeth; Ju, Wendy; Soh, Harold; Williams, Tom (Ed.): Companion of the 2024 ACM/IEEE International Conference on Human-Robot Interaction, HRI 2024, Boulder, CO, USA, March 11-15, 2024, pp. 332–336, ACM, 2024.
@inproceedings{DBLP:conf/hri/ChristofiB24,
title = {Uncovering Patterns in Humans that Teach Robots through Demonstrations
and Feedback},
author = {Konstantinos Christofi and
Kim Baraka},
editor = {Dan Grollman and
Elizabeth Broadbent and
Wendy Ju and
Harold Soh and
Tom Williams},
url = {https://doi.org/10.1145/3610978.3640740},
doi = {10.1145/3610978.3640740},
year = {2024},
date = {2024-01-01},
booktitle = {Companion of the 2024 ACM/IEEE International Conference on Human-Robot
Interaction, HRI 2024, Boulder, CO, USA, March 11-15, 2024},
pages = {332–336},
publisher = {ACM},
abstract = {Human-in-the-loop robot learning allows a robot to learn tasks more effectively with the help of humans in the role of teacher. While there is a large body of work on algorithms that leverage human input for better robot learning, there has been little attention to understanding how humans teach robots. In this paper, we provide preliminary results on how users strategize the use of demonstrations and evaluative feedback under a budget, and how these choices are influenced by demographic variables such as gender. We implemented a learning algorithm that allows a simulated robot arm to learn three reaching tasks with the help of a human. We collected interaction data for a total of 58 participants, which shows that participants demonstrate a tendency to provide evaluative feedback earlier in their interactions compared to demonstrations, and that gender may have an influence on teaching strategy. This preliminary analysis lays the foundation for future research aimed at developing tuneable computational models of different human teachers.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Human-in-the-loop robot learning allows a robot to learn tasks more effectively with the help of humans in the role of teacher. While there is a large body of work on algorithms that leverage human input for better robot learning, there has been little attention to understanding how humans teach robots. In this paper, we provide preliminary results on how users strategize the use of demonstrations and evaluative feedback under a budget, and how these choices are influenced by demographic variables such as gender. We implemented a learning algorithm that allows a simulated robot arm to learn three reaching tasks with the help of a human. We collected interaction data for a total of 58 participants, which shows that participants demonstrate a tendency to provide evaluative feedback earlier in their interactions compared to demonstrations, and that gender may have an influence on teaching strategy. This preliminary analysis lays the foundation for future research aimed at developing tuneable computational models of different human teachers.
185.
Mandal A; Baraka K
Using Proxemics as a Corrective Feedback Signal during Robot Navigation Proceedings Article
In: Grollman, Dan; Broadbent, Elizabeth; Ju, Wendy; Soh, Harold; Williams, Tom (Ed.): Companion of the 2024 ACM/IEEE International Conference on Human-Robot Interaction, HRI 2024, Boulder, CO, USA, March 11-15, 2024, pp. 732–736, ACM, 2024.
@inproceedings{DBLP:conf/hri/MandalB24,
title = {Using Proxemics as a Corrective Feedback Signal during Robot Navigation},
author = {Adwitiya Mandal and
Kim Baraka},
editor = {Dan Grollman and
Elizabeth Broadbent and
Wendy Ju and
Harold Soh and
Tom Williams},
url = {https://doi.org/10.1145/3610978.3640746},
doi = {10.1145/3610978.3640746},
year = {2024},
date = {2024-01-01},
booktitle = {Companion of the 2024 ACM/IEEE International Conference on Human-Robot
Interaction, HRI 2024, Boulder, CO, USA, March 11-15, 2024},
pages = {732–736},
publisher = {ACM},
abstract = {This paper presents a novel feedback system based on proxemics. PFbS (Proximity Feedback System) allows a robot to dynamically update its navigational behavior by interpreting a human teacher's distance to the robot as a corrective feedback signal. We demonstrate PFbS on a Pepper robot and evaluate it in a pilot study (N=18) against a joystick interface baseline, in terms of usability, intuitiveness and signal clarity. We also evaluate the learning algorithm for dynamic update of the robot's trajectory. Based on the insights from the pilot, we believe that with proper sensor accuracy, PFbS can be improved to allow for more intuitive and embodied feedback for non-expert users.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
This paper presents a novel feedback system based on proxemics. PFbS (Proximity Feedback System) allows a robot to dynamically update its navigational behavior by interpreting a human teacher's distance to the robot as a corrective feedback signal. We demonstrate PFbS on a Pepper robot and evaluate it in a pilot study (N=18) against a joystick interface baseline, in terms of usability, intuitiveness and signal clarity. We also evaluate the learning algorithm for dynamic update of the robot's trajectory. Based on the insights from the pilot, we believe that with proper sensor accuracy, PFbS can be improved to allow for more intuitive and embodied feedback for non-expert users.
184.
Schrage-Prent N; Vanegas D F P; Baraka K
Interactive Robot Programming Inspired by Dog Training: An Exploratory Study Proceedings Article
In: Grollman, Dan; Broadbent, Elizabeth; Ju, Wendy; Soh, Harold; Williams, Tom (Ed.): Companion of the 2024 ACM/IEEE International Conference on Human-Robot Interaction, HRI 2024, Boulder, CO, USA, March 11-15, 2024, pp. 965–969, ACM, 2024.
@inproceedings{DBLP:conf/hri/Schrage-PrentVB24,
title = {Interactive Robot Programming Inspired by Dog Training: An Exploratory
Study},
author = {Nienke Schrage{-}Prent and
Daniel F. Preciado Vanegas and
Kim Baraka},
editor = {Dan Grollman and
Elizabeth Broadbent and
Wendy Ju and
Harold Soh and
Tom Williams},
url = {https://doi.org/10.1145/3610978.3640655},
doi = {10.1145/3610978.3640655},
year = {2024},
date = {2024-01-01},
booktitle = {Companion of the 2024 ACM/IEEE International Conference on Human-Robot
Interaction, HRI 2024, Boulder, CO, USA, March 11-15, 2024},
pages = {965–969},
publisher = {ACM},
abstract = {Programming a robot takes time, effort, and expert knowledge. As robots find their way to our personal spaces, it becomes urgent to investigate more intuitive methods to program them. An emerging field of research has focused on developing systems that are easy for non-expert users to understand and train. This paper explores the premise that dog training methods could inspire interactive programming methods for robots. In collaboration with dog trainers, we designed an interactive programming method for a robot. We evaluated our method in a Wizard-of-Oz study with 18 participants and compared it with programming the same behavior on a graphical programming software. Results show significant differences in usability scores, with the method inspired by dog training being perceived as more usable, easier, more fun, and more personal. This suggests that robot programming methods based on dog training could benefit a broader range of end-users, allowing them to interactively program new behaviors on robots in richer and more intuitive ways.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Programming a robot takes time, effort, and expert knowledge. As robots find their way to our personal spaces, it becomes urgent to investigate more intuitive methods to program them. An emerging field of research has focused on developing systems that are easy for non-expert users to understand and train. This paper explores the premise that dog training methods could inspire interactive programming methods for robots. In collaboration with dog trainers, we designed an interactive programming method for a robot. We evaluated our method in a Wizard-of-Oz study with 18 participants and compared it with programming the same behavior on a graphical programming software. Results show significant differences in usability scores, with the method inspired by dog training being perceived as more usable, easier, more fun, and more personal. This suggests that robot programming methods based on dog training could benefit a broader range of end-users, allowing them to interactively program new behaviors on robots in richer and more intuitive ways.
183.
Knierim M; Jain S; Aydogan M H; Mitra K; Desai K; Saran A; Baraka K
Leveraging Prosody as an Informative Teaching Signal for Agent Learning: Exploratory Studies and Algorithmic Implications Proceedings Article
In: Hung, Hayley; Oertel, Catharine; Soleymani, Mohammad; Chaspari, Theodora; Dibeklioglu, Hamdi; Shukla, Jainendra; Truong, Khiet P. (Ed.): Proceedings of the 26th International Conference on Multimodal Interaction, ICMI 2024, San Jose, Costa Rica, November 4-8, 2024, pp. 95–123, ACM, 2024.
@inproceedings{DBLP:conf/icmi/KnierimJAMDSB24,
title = {Leveraging Prosody as an Informative Teaching Signal for Agent Learning:
Exploratory Studies and Algorithmic Implications},
author = {Matilda Knierim and
Sahil Jain and
Murat Han Aydogan and
Kenneth Mitra and
Kush Desai and
Akanksha Saran and
Kim Baraka},
editor = {Hayley Hung and
Catharine Oertel and
Mohammad Soleymani and
Theodora Chaspari and
Hamdi Dibeklioglu and
Jainendra Shukla and
Khiet P. Truong},
url = {https://doi.org/10.1145/3678957.3685735},
doi = {10.1145/3678957.3685735},
year = {2024},
date = {2024-01-01},
booktitle = {Proceedings of the 26th International Conference on Multimodal Interaction,
ICMI 2024, San Jose, Costa Rica, November 4-8, 2024},
pages = {95–123},
publisher = {ACM},
abstract = {Agent learning from human interaction often relies on explicit signals, but implicit social cues, such as prosody in speech, could provide valuable information for more effective learning. This paper advocates for the integration of prosody as a teaching signal to enhance agent learning from human teachers. Through two exploratory studies—one examining voice feedback in an interactive reinforcement learning setup and the other analyzing restricted audio from human demonstrations in three Atari games—we demonstrate that prosody carries significant information about task dynamics. Our findings suggest that prosodic features, when coupled with explicit feedback, can enhance reinforcement learning outcomes. Moreover, we propose guidelines for prosody-sensitive algorithm design and discuss insights into teaching behavior. Our work underscores the potential of leveraging prosody as an implicit signal for more efficient agent learning, thus advancing human-agent interaction paradigms.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Agent learning from human interaction often relies on explicit signals, but implicit social cues, such as prosody in speech, could provide valuable information for more effective learning. This paper advocates for the integration of prosody as a teaching signal to enhance agent learning from human teachers. Through two exploratory studies—one examining voice feedback in an interactive reinforcement learning setup and the other analyzing restricted audio from human demonstrations in three Atari games—we demonstrate that prosody carries significant information about task dynamics. Our findings suggest that prosodic features, when coupled with explicit feedback, can enhance reinforcement learning outcomes. Moreover, we propose guidelines for prosody-sensitive algorithm design and discuss insights into teaching behavior. Our work underscores the potential of leveraging prosody as an implicit signal for more efficient agent learning, thus advancing human-agent interaction paradigms.
182.
Adamik M; Pernisch R; Tiddi I; Schlobach S
ORKA: An Ontology for Robotic Knowledge Acquisition Proceedings Article
In: Alam, Mehwish; Rospocher, Marco; van Erp, Marieke; Hollink, Laura; Gesese, Genet Asefa (Ed.): Knowledge Engineering and Knowledge Management - 24th International Conference, EKAW 2024, Amsterdam, The Netherlands, November 26-28, 2024, Proceedings, pp. 309–327, Springer, 2024.
@inproceedings{DBLP:conf/ekaw/AdamikPTS24,
title = {ORKA: An Ontology for Robotic Knowledge Acquisition},
author = {Mark Adamik and
Romana Pernisch and
Ilaria Tiddi and
Stefan Schlobach},
editor = {Mehwish Alam and
Marco Rospocher and
Marieke van Erp and
Laura Hollink and
Genet Asefa Gesese},
url = {https://doi.org/10.1007/978-3-031-77792-9_19},
doi = {10.1007/978-3-031-77792-9_19},
year = {2024},
date = {2024-01-01},
booktitle = {Knowledge Engineering and Knowledge Management - 24th International
Conference, EKAW 2024, Amsterdam, The Netherlands, November 26-28,
2024, Proceedings},
volume = {15370},
pages = {309–327},
publisher = {Springer},
series = {Lecture Notes in Computer Science},
abstract = {Most autonomous agents operating in the real world use perception capabilities and reasoning mechanisms to acquire new knowledge of the environment, where perception capabilities include both the physical sensor devices and the software-based perception pipelines involved in the process. For autonomous agents to be able to adjust and reason over their own perception, knowledge of the sensors and the corresponding perception algorithms is required. We present the Ontology for Robotic Knowledge Acquisition (ORKA), that models the perception pipeline of a robotic agent by representing the sensory, algorithmic and measurement aspects of the perception process, thereby unifying the agent’s sensing with the characteristics of the environment and facilitating the grounding process. The ontology is based on the alignment between SSN and OBOE, linked to external databases as additional knowledge sources for robotic agents, populated wit instances from two different robotic use-cases, and evaluated using competency questions and comparisons to related ontologies. A proof of concept use-case is presented to highlight the potential of the ontology.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Most autonomous agents operating in the real world use perception capabilities and reasoning mechanisms to acquire new knowledge of the environment, where perception capabilities include both the physical sensor devices and the software-based perception pipelines involved in the process. For autonomous agents to be able to adjust and reason over their own perception, knowledge of the sensors and the corresponding perception algorithms is required. We present the Ontology for Robotic Knowledge Acquisition (ORKA), that models the perception pipeline of a robotic agent by representing the sensory, algorithmic and measurement aspects of the perception process, thereby unifying the agent’s sensing with the characteristics of the environment and facilitating the grounding process. The ontology is based on the alignment between SSN and OBOE, linked to external databases as additional knowledge sources for robotic agents, populated wit instances from two different robotic use-cases, and evaluated using competency questions and comparisons to related ontologies. A proof of concept use-case is presented to highlight the potential of the ontology.
181.
Adamik M; Pernisch R; Tiddi I; Schlobach S
Advancing Robotic Perception with Perceived-Entity Linking Proceedings Article
In: Demartini, Gianluca; Hose, Katja; Acosta, Maribel; Palmonari, Matteo; Cheng, Gong; Skaf-Molli, Hala; Ferranti, Nicolas; Hernández, Daniel; Hogan, Aidan (Ed.): The Semantic Web - ISWC 2024 - 23rd International Semantic Web Conference, Baltimore, MD, USA, November 11-15, 2024, Proceedings, Part II, pp. 192–209, Springer, 2024.
@inproceedings{DBLP:conf/semweb/AdamikPTS24,
title = {Advancing Robotic Perception with Perceived-Entity Linking},
author = {Mark Adamik and
Romana Pernisch and
Ilaria Tiddi and
Stefan Schlobach},
editor = {Gianluca Demartini and
Katja Hose and
Maribel Acosta and
Matteo Palmonari and
Gong Cheng and
Hala Skaf{-}Molli and
Nicolas Ferranti and
Daniel Hernández and
Aidan Hogan},
url = {https://doi.org/10.1007/978-3-031-77850-6_11},
doi = {10.1007/978-3-031-77850-6_11},
year = {2024},
date = {2024-01-01},
booktitle = {The Semantic Web - ISWC 2024 - 23rd International Semantic Web Conference,
Baltimore, MD, USA, November 11-15, 2024, Proceedings, Part II},
volume = {15232},
pages = {192–209},
publisher = {Springer},
series = {Lecture Notes in Computer Science},
abstract = {The capabilities of current robotic applications are significantly constrained by their limited ability to perceive and understand their surroundings. The Semantic Web aims to offer general, machine-readable knowledge about the world and could be a potential solution to address the information needs of robotic agents. We introduce the Perceived-Entity Linking (PEL) problem as the task of recognizing entities and linking the sensory data of an autonomous agent to a unique identifier in a target knowledge graph. We provide a formal definition of PEL, and propose a PEL baseline based on the YOLO object detection algorithm and a conventional entity linking method as an initial attempt to solve the task. The baseline is evaluated by linking the concepts contained in MS COCO and VisualGenome datasets to WikiData, DBpedia and YAGO as target knowledge graphs. This study makes a first step in allowing robotic agents to leverage the extensive knowledge contained in general-purpose knowledge graphs},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
The capabilities of current robotic applications are significantly constrained by their limited ability to perceive and understand their surroundings. The Semantic Web aims to offer general, machine-readable knowledge about the world and could be a potential solution to address the information needs of robotic agents. We introduce the Perceived-Entity Linking (PEL) problem as the task of recognizing entities and linking the sensory data of an autonomous agent to a unique identifier in a target knowledge graph. We provide a formal definition of PEL, and propose a PEL baseline based on the YOLO object detection algorithm and a conventional entity linking method as an initial attempt to solve the task. The baseline is evaluated by linking the concepts contained in MS COCO and VisualGenome datasets to WikiData, DBpedia and YAGO as target knowledge graphs. This study makes a first step in allowing robotic agents to leverage the extensive knowledge contained in general-purpose knowledge graphs
180.
Zamprogno G; Adamik M; Roothaert R; Naghdipour A; Stork L; Koopmann P; Pernisch R; Kruit B; Chen J; Tiddi I; Schlobach S
Supporting Companion Planting with the CoPla Ontology Proceedings Article
In: Blomqvist, Eva; Garc'ıa-Castro, Raúl; Hernández, Daniel; Hitzler, Pascal; Lindecrantz, Mikael; Poveda-Villalón, Mar'ıa (Ed.): Proceedings of the The 2nd International Workshop on Knowledge Graphs for Sustainability (KG4S 2024) colocated with the 21st Extended Semantic Web Conference (ESWC 2024), Hersonissos, Greece, May 27th, 2024, pp. 29–41, CEUR-WS.org, 2024.
@inproceedings{DBLP:conf/kg4s/ZamprognoARNSKP24,
title = {Supporting Companion Planting with the CoPla Ontology},
author = {Giacomo Zamprogno and
Mark Adamik and
Ritten Roothaert and
Ameneh Naghdipour and
Lise Stork and
Patrick Koopmann and
Romana Pernisch and
Benno Kruit and
Jieying Chen and
Ilaria Tiddi and
Stefan Schlobach},
editor = {Eva Blomqvist and
Ra{ú}l Garc{'{ı}}a{-}Castro and
Daniel Hernández and
Pascal Hitzler and
Mikael Lindecrantz and
Mar{'{ı}}a Poveda{-}Villalón},
url = {https://ceur-ws.org/Vol-3753/paper3.pdf},
year = {2024},
date = {2024-01-01},
booktitle = {Proceedings of the The 2nd International Workshop on Knowledge Graphs
for Sustainability (KG4S 2024) colocated with the 21st Extended
Semantic Web Conference (ESWC 2024), Hersonissos, Greece, May 27th,
2024},
volume = {3753},
pages = {29–41},
publisher = {CEUR-WS.org},
series = {CEUR Workshop Proceedings},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
179.
Chen H; Kuzina A; Esmaeili B; Tomczak J M
Variational Stochastic Gradient Descent for Deep Neural Networks Journal Article
In: International Conference on Machine Learning. Workshop on Advancing Neural Network Training, 2024.
@article{chen2024variational,
title = {Variational Stochastic Gradient Descent for Deep Neural Networks},
author = {Chen, Haotian and Kuzina, Anna and Esmaeili, Babak and Tomczak, Jakub M},
url = {https://openreview.net/forum?id=QLo5lGkiyg},
year = {2024},
date = {2024-01-01},
journal = {International Conference on Machine Learning. Workshop on Advancing Neural Network Training},
abstract = {Optimizing deep neural networks (DNNs) is one of the main tasks in successful deep learning. Current state-of-the-art optimizers are adaptive gradient-based optimization methods such as Adam. Recently, there has been an increasing interest in formulating gradient-based optimizers in a probabilistic framework for better estimation of gradients and modeling uncertainties. Here, we propose to combine both approaches, resulting in the Variational Stochastic Gradient Descent (VSGD) optimizer. We model gradient updates as a probabilistic model and utilize stochastic variational inference (SVI) to derive an efficient and effective update rule. Further, we show how our VSGD method relates to other adaptive gradient-based optimizers like Adam. Lastly, we carry out experiments on two image classification datasets and three deep neural network architectures, where we show that VSGD converges faster and outperforms Adam and SGD.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Optimizing deep neural networks (DNNs) is one of the main tasks in successful deep learning. Current state-of-the-art optimizers are adaptive gradient-based optimization methods such as Adam. Recently, there has been an increasing interest in formulating gradient-based optimizers in a probabilistic framework for better estimation of gradients and modeling uncertainties. Here, we propose to combine both approaches, resulting in the Variational Stochastic Gradient Descent (VSGD) optimizer. We model gradient updates as a probabilistic model and utilize stochastic variational inference (SVI) to derive an efficient and effective update rule. Further, we show how our VSGD method relates to other adaptive gradient-based optimizers like Adam. Lastly, we carry out experiments on two image classification datasets and three deep neural network architectures, where we show that VSGD converges faster and outperforms Adam and SGD.
178.
Kuzina A; Tomczak J M
Hierarchical VAE with a Diffusion-based VampPrior Journal Article
In: Transactions on Machine Learning Research, 2024.
@article{kuzina2024hierarchical,
title = {Hierarchical VAE with a Diffusion-based VampPrior},
author = {Kuzina, Anna and Tomczak, Jakub M},
url = {https://openreview.net/forum?id=NUkEoZ7Toa},
year = {2024},
date = {2024-01-01},
journal = {Transactions on Machine Learning Research},
abstract = {Deep hierarchical variational autoencoders (VAEs) are powerful latent variable generative models. In this paper, we introduce Hierarchical VAE with Diffusion-based Variational Mixture of the Posterior Prior (VampPrior). We apply amortization to scale the VampPrior to models with many stochastic layers. The proposed approach allows us to achieve better performance compared to the original VampPrior work and other deep hierarchical VAEs, while using fewer parameters. We empirically validate our method on standard benchmark datasets (MNIST, OMNIGLOT, CIFAR10) and demonstrate improved training stability and latent space utilization.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Deep hierarchical variational autoencoders (VAEs) are powerful latent variable generative models. In this paper, we introduce Hierarchical VAE with Diffusion-based Variational Mixture of the Posterior Prior (VampPrior). We apply amortization to scale the VampPrior to models with many stochastic layers. The proposed approach allows us to achieve better performance compared to the original VampPrior work and other deep hierarchical VAEs, while using fewer parameters. We empirically validate our method on standard benchmark datasets (MNIST, OMNIGLOT, CIFAR10) and demonstrate improved training stability and latent space utilization.
177.
Airiau S; Dupuis N K; Grossi D
Condorcet Markets Proceedings Article
In: Schäfer, Guido; Ventre, Carmine (Ed.): Algorithmic Game Theory - 17th International Symposium, SAGT 2024, Amsterdam, The Netherlands, September 3-6, 2024, Proceedings, pp. 501–519, Springer, 2024.
@inproceedings{DBLP:conf/sagt/AiriauDG24,
title = {Condorcet Markets},
author = {Stéphane Airiau and Nicholas Kees Dupuis and Davide Grossi},
editor = {Guido Schäfer and Carmine Ventre},
url = {https://doi.org/10.1007/978-3-031-71033-9_28},
doi = {10.1007/978-3-031-71033-9_28},
year = {2024},
date = {2024-01-01},
booktitle = {Algorithmic Game Theory - 17th International Symposium, SAGT 2024, Amsterdam, The Netherlands, September 3-6, 2024, Proceedings},
volume = {15156},
pages = {501–519},
publisher = {Springer},
series = {Lecture Notes in Computer Science},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
176.
Venema-Los M; Christoff Z; Grossi D
Limited Voting for Better Representation? Proceedings Article
In: Endriss, Ulle; Melo, Francisco S.; Bach, Kerstin; Diz, Alberto José Bugar'ın; Alonso-Moral, Jose Maria; Barro, Senén; Heintz, Fredrik (Ed.): ECAI 2024 - 27th European Conference on Artificial Intelligence, 19-24 October 2024, Santiago de Compostela, Spain - Including 13th Conference on Prestigious Applications of Intelligent Systems (PAIS 2024), pp. 3509–3516, IOS Press, 2024.
@inproceedings{DBLP:conf/ecai/Venema-LosCG24,
title = {Limited Voting for Better Representation?},
author = {Maaike Venema{-}Los and Zoé Christoff and Davide Grossi},
editor = {Ulle Endriss and Francisco S. Melo and Kerstin Bach and Alberto José Bugar{'{ı}}n Diz and Jose Maria Alonso{-}Moral and Senén Barro and Fredrik Heintz},
url = {https://doi.org/10.3233/FAIA240904},
doi = {10.3233/FAIA240904},
year = {2024},
date = {2024-01-01},
booktitle = {ECAI 2024 - 27th European Conference on Artificial Intelligence, 19-24 October 2024, Santiago de Compostela, Spain - Including 13th Conference on Prestigious Applications of Intelligent Systems (PAIS 2024)},
volume = {392},
pages = {3509–3516},
publisher = {IOS Press},
series = {Frontiers in Artificial Intelligence and Applications},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
175.
Orzan N; Acar E; Grossi D; Mannion P; Radulescu R
Learning in Multi-Objective Public Goods Games with Non-Linear Utilities Proceedings Article
In: Endriss, Ulle; Melo, Francisco S.; Bach, Kerstin; Diz, Alberto José Bugar'ın; Alonso-Moral, Jose Maria; Barro, Senén; Heintz, Fredrik (Ed.): ECAI 2024 - 27th European Conference on Artificial Intelligence, 19-24 October 2024, Santiago de Compostela, Spain - Including 13th Conference on Prestigious Applications of Intelligent Systems (PAIS 2024), pp. 2749–2756, IOS Press, 2024.
@inproceedings{DBLP:conf/ecai/OrzanAGMR24,
title = {Learning in Multi-Objective Public Goods Games with Non-Linear Utilities},
author = {Nicole Orzan and Erman Acar and Davide Grossi and Patrick Mannion and Roxana Radulescu},
editor = {Ulle Endriss and Francisco S. Melo and Kerstin Bach and Alberto José Bugar{'{ı}}n Diz and Jose Maria Alonso{-}Moral and Senén Barro and Fredrik Heintz},
url = {https://doi.org/10.3233/FAIA240809},
doi = {10.3233/FAIA240809},
year = {2024},
date = {2024-01-01},
booktitle = {ECAI 2024 - 27th European Conference on Artificial Intelligence, 19-24 October 2024, Santiago de Compostela, Spain - Including 13th Conference on Prestigious Applications of Intelligent Systems (PAIS 2024)},
volume = {392},
pages = {2749–2756},
publisher = {IOS Press},
series = {Frontiers in Artificial Intelligence and Applications},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
174.
Orzan N; Acar E; Grossi D; Radulescu R
Emergent Cooperation under Uncertain Incentive Alignment Proceedings Article
In: Dastani, Mehdi; Sichman, Jaime Simão; Alechina, Natasha; Dignum, Virginia (Ed.): Proceedings of the 23rd International Conference on Autonomous Agents and Multiagent Systems, AAMAS 2024, Auckland, New Zealand, May 6-10, 2024, pp. 1521–1530, International Foundation for Autonomous Agents and Multiagent Systems / ACM, 2024.
@inproceedings{DBLP:conf/atal/OrzanAGR24,
title = {Emergent Cooperation under Uncertain Incentive Alignment},
author = {Nicole Orzan and Erman Acar and Davide Grossi and Roxana Radulescu},
editor = {Mehdi Dastani and Jaime Sim{ã}o Sichman and Natasha Alechina and Virginia Dignum},
url = {https://dl.acm.org/doi/10.5555/3635637.3663012},
doi = {10.5555/3635637.3663012},
year = {2024},
date = {2024-01-01},
booktitle = {Proceedings of the 23rd International Conference on Autonomous Agents and Multiagent Systems, AAMAS 2024, Auckland, New Zealand, May 6-10, 2024},
pages = {1521–1530},
publisher = {International Foundation for Autonomous Agents and Multiagent Systems / ACM},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
173.
Dell'Anna D; Murukannaiah P K; Dudzik B; Grossi D; Jonker C M; Oertel C; Yolum P
Toward a Quality Model for Hybrid Intelligence Teams Proceedings Article
In: Dastani, Mehdi; Sichman, Jaime Simão; Alechina, Natasha; Dignum, Virginia (Ed.): Proceedings of the 23rd International Conference on Autonomous Agents and Multiagent Systems, AAMAS 2024, Auckland, New Zealand, May 6-10, 2024, pp. 434–443, International Foundation for Autonomous Agents and Multiagent Systems / ACM, 2024.
@inproceedings{DBLP:conf/atal/DellAnnaMDGJOY24,
title = {Toward a Quality Model for Hybrid Intelligence Teams},
author = {Davide Dell'Anna and Pradeep K. Murukannaiah and Bernd Dudzik and Davide Grossi and Catholijn M. Jonker and Catharine Oertel and Pinar Yolum},
editor = {Mehdi Dastani and Jaime Sim{ã}o Sichman and Natasha Alechina and Virginia Dignum},
url = {https://dl.acm.org/doi/10.5555/3635637.3662893},
doi = {10.5555/3635637.3662893},
year = {2024},
date = {2024-01-01},
booktitle = {Proceedings of the 23rd International Conference on Autonomous Agents and Multiagent Systems, AAMAS 2024, Auckland, New Zealand, May 6-10, 2024},
pages = {434–443},
publisher = {International Foundation for Autonomous Agents and Multiagent Systems / ACM},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
172.
Elkind E; Grossi D; Shapiro E; Talmon N
United for change: deliberative coalition formation to change the status quo Journal Article
In: Soc. Choice Welf., vol. 63, no. 3, pp. 717–746, 2024.
@article{DBLP:journals/scw/ElkindGST24,
title = {United for change: deliberative coalition formation to change the status quo},
author = {Edith Elkind and Davide Grossi and Ehud Shapiro and Nimrod Talmon},
url = {https://doi.org/10.1007/s00355-024-01561-y},
doi = {10.1007/S00355-024-01561-Y},
year = {2024},
date = {2024-01-01},
journal = {Soc. Choice Welf.},
volume = {63},
number = {3},
pages = {717–746},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
171.
van Woerkom W; Grossi D; Prakken H; Verheij B
A Fortiori Case-Based Reasoning: From Theory to Data Journal Article
In: J. Artif. Intell. Res., vol. 81, pp. 401–441, 2024.
@article{DBLP:journals/jair/WoerkomGPV24,
title = {A Fortiori Case-Based Reasoning: From Theory to Data},
author = {Wijnand van Woerkom and Davide Grossi and Henry Prakken and Bart Verheij},
url = {https://doi.org/10.1613/jair.1.15178},
doi = {10.1613/JAIR.1.15178},
year = {2024},
date = {2024-01-01},
journal = {J. Artif. Intell. Res.},
volume = {81},
pages = {401–441},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
170.
Garc'ıa-Castellanos A; Medbouhi A A; Marchetti G L; Bekkers E J; Kragic D
HyperSteiner: Computing Heuristic Hyperbolic Steiner Minimal Trees Journal Article
In: vol. abs/2409.05671, 2024.
@article{DBLP:journals/corr/abs-2409-05671,
title = {HyperSteiner: Computing Heuristic Hyperbolic Steiner Minimal Trees},
author = {Alejandro Garc{'{ı}}a{-}Castellanos and
Aniss Aiman Medbouhi and
Giovanni Luca Marchetti and
Erik J. Bekkers and
Danica Kragic},
url = {https://arxiv.org/abs/2409.05671},
doi = {10.48550/ARXIV.2409.05671},
year = {2024},
date = {2024-01-01},
volume = {abs/2409.05671},
abstract = {We propose HyperSteiner – an efficient heuristic algorithm for computing Steiner minimal trees in the hyperbolic space. HyperSteiner extends the Euclidean Smith-Lee-Liebman algorithm, which is grounded in a divide-and-conquer approach involving the Delaunay triangulation. The central idea is rephrasing Steiner tree problems with three terminals as a system of equations in the Klein-Beltrami model. Motivated by the fact that hyperbolic geometry is well-suited for representing hierarchies, we explore applications to hierarchy discovery in data. Results show that HyperSteiner infers more realistic hierarchies than the Minimum Spanning Tree and is more scalable to large datasets than Neighbor Joining. journal = CoRR},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
We propose HyperSteiner – an efficient heuristic algorithm for computing Steiner minimal trees in the hyperbolic space. HyperSteiner extends the Euclidean Smith-Lee-Liebman algorithm, which is grounded in a divide-and-conquer approach involving the Delaunay triangulation. The central idea is rephrasing Steiner tree problems with three terminals as a system of equations in the Klein-Beltrami model. Motivated by the fact that hyperbolic geometry is well-suited for representing hierarchies, we explore applications to hierarchy discovery in data. Results show that HyperSteiner infers more realistic hierarchies than the Minimum Spanning Tree and is more scalable to large datasets than Neighbor Joining. journal = CoRR
169.
Garc'ıa-Castellanos A; Marchetti G L; Kragic D; Scolamiero M
Relative Representations: Topological and Geometric Perspectives Journal Article
In: vol. abs/2409.10967, 2024.
@article{DBLP:journals/corr/abs-2409-10967,
title = {Relative Representations: Topological and Geometric Perspectives},
author = {Alejandro Garc{'{ı}}a{-}Castellanos and
Giovanni Luca Marchetti and
Danica Kragic and
Martina Scolamiero},
url = {https://arxiv.org/abs/2409.10967},
doi = {10.48550/ARXIV.2409.10967},
year = {2024},
date = {2024-01-01},
volume = {abs/2409.10967},
abstract = {Relative representations are an established approach to zero-shot model stitching, consisting of a non-trainable transformation of the latent space of a deep neural network. Based on insights of topological and geometric nature, we propose two improvements to relative representations. First, we introduce a normalization procedure in the relative transformation, resulting in invariance to non-isotropic rescalings and permutations. The latter coincides with the symmetries in parameter space induced by common activation functions. Second, we propose to deploy topological densification when fine-tuning relative representations, a topological regularization loss encouraging clustering within classes. We provide an empirical investigation on a natural language task, where both the proposed variations yield improved performance on zero-shot model stitching. journal = CoRR},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Relative representations are an established approach to zero-shot model stitching, consisting of a non-trainable transformation of the latent space of a deep neural network. Based on insights of topological and geometric nature, we propose two improvements to relative representations. First, we introduce a normalization procedure in the relative transformation, resulting in invariance to non-isotropic rescalings and permutations. The latter coincides with the symmetries in parameter space induced by common activation functions. Second, we propose to deploy topological densification when fine-tuning relative representations, a topological regularization loss encouraging clustering within classes. We provide an empirical investigation on a natural language task, where both the proposed variations yield improved performance on zero-shot model stitching. journal = CoRR
168.
de Boer V; Stork L
Hybrid Intelligence for Digital Humanities Book Section
In: HHAI 2024: Hybrid Human AI Systems for the Social Good, pp. 94–104, IOS Press, 2024.
@incollection{de2024hybrid,
title = {Hybrid Intelligence for Digital Humanities},
author = {de Boer, Victor and Stork, Lise},
year = {2024},
date = {2024-01-01},
booktitle = {HHAI 2024: Hybrid Human AI Systems for the Social Good},
pages = {94–104},
publisher = {IOS Press},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
167.
Wolff J; De Boer V; Heylen D; Van Riemsdijk M B
Using Default Logic to Create Adaptable User Models for Behavior Support Agents Book Section
In: HHAI 2024: Hybrid Human AI Systems for the Social Good, pp. 350–359, IOS Press, 2024.
@incollection{wolff2024using,
title = {Using Default Logic to Create Adaptable User Models for Behavior Support Agents},
author = {Wolff, Johanna and De Boer, Victor and Heylen, Dirk and Van Riemsdijk, M Birna},
year = {2024},
date = {2024-01-01},
booktitle = {HHAI 2024: Hybrid Human AI Systems for the Social Good},
pages = {350–359},
publisher = {IOS Press},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
166.
Wolff J; de Boer V; Heylen D; van Riemsdijk M B
Defining an Adaptable Framework for Behaviour Support Agents in Default Logic Proceedings Article
In: 22nd International Workshop on Nonmonotonic Reasoning, NMR 2024, pp. 72–82, CEUR-WS 2024.
@inproceedings{wolff2024defining,
title = {Defining an Adaptable Framework for Behaviour Support Agents in Default Logic},
author = {Wolff, Johanna and de Boer, Victor and Heylen, Dirk and van Riemsdijk, M Birna},
year = {2024},
date = {2024-01-01},
booktitle = {22nd International Workshop on Nonmonotonic Reasoning, NMR 2024},
pages = {72–82},
organization = {CEUR-WS},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
165.
Loftin R; Bandyopadhyay S; Çelikok M M
On the Complexity of Learning to Cooperate with Populations of Socially Rational Agents Journal Article
In: arXiv preprint arXiv:2407.00419, 2024.
@article{loftin2024complexity,
title = {On the Complexity of Learning to Cooperate with Populations of Socially Rational Agents},
author = {Loftin, Robert and Bandyopadhyay, Saptarashmi and {Ç}elikok, Mustafa Mert},
year = {2024},
date = {2024-01-01},
urldate = {2024-01-01},
journal = {arXiv preprint arXiv:2407.00419},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
164.
van Dijk B; van Duijn M; Verberne S; Spruit M
ChiSCor: A Corpus of Freely-Told Fantasy Stories by Dutch Children for Computational Linguistics and Cognitive Science Proceedings Article
In: Jiang, Jing; Reitter, David; Deng, Shumin (Ed.): Proceedings of the 27th Conference on Computational Natural Language Learning (CoNLL), pp. 352–363, Association for Computational Linguistics, Singapore, 2023.
@inproceedings{van-dijk-etal-2023-chiscor,
title = {ChiSCor: A Corpus of Freely-Told Fantasy Stories by Dutch Children for Computational Linguistics and Cognitive Science},
author = {van Dijk, Bram and
van Duijn, Max and
Verberne, Suzan and
Spruit, Marco},
editor = {Jiang, Jing and
Reitter, David and
Deng, Shumin},
url = {https://aclanthology.org/2023.conll-1.23},
doi = {10.18653/v1/2023.conll-1.23},
year = {2023},
date = {2023-12-01},
booktitle = {Proceedings of the 27th Conference on Computational Natural Language Learning (CoNLL)},
pages = {352–363},
publisher = {Association for Computational Linguistics},
address = {Singapore},
abstract = {In this resource paper we release ChiSCor, a new corpus containing 619 fantasy stories, told freely by 442 Dutch children aged 4-12. ChiSCor was compiled for studying how children render character perspectives, and unravelling language and cognition in development, with computational tools. Unlike existing resources, ChiSCor's stories were produced in natural contexts, in line with recent calls for more ecologically valid datasets. ChiSCor hosts text, audio, and annotations for character complexity and linguistic complexity. Additional metadata (e.g. education of caregivers) is available for one third of the Dutch children. ChiSCor also includes a small set of 62 English stories. This paper details how ChiSCor was compiled and shows its potential for future work with three brief case studies: i) we show that the syntactic complexity of stories is strikingly stable across children's ages; ii) we extend work on Zipfian distributions in free speech and show that ChiSCor obeys Zipf's law closely, reflecting its social context; iii) we show that even though ChiSCor is relatively small, the corpus is rich enough to train informative lemma vectors that allow us to analyse children's language use. We end with a reflection on the value of narrative datasets in computational linguistics.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
In this resource paper we release ChiSCor, a new corpus containing 619 fantasy stories, told freely by 442 Dutch children aged 4-12. ChiSCor was compiled for studying how children render character perspectives, and unravelling language and cognition in development, with computational tools. Unlike existing resources, ChiSCor's stories were produced in natural contexts, in line with recent calls for more ecologically valid datasets. ChiSCor hosts text, audio, and annotations for character complexity and linguistic complexity. Additional metadata (e.g. education of caregivers) is available for one third of the Dutch children. ChiSCor also includes a small set of 62 English stories. This paper details how ChiSCor was compiled and shows its potential for future work with three brief case studies: i) we show that the syntactic complexity of stories is strikingly stable across children's ages; ii) we extend work on Zipfian distributions in free speech and show that ChiSCor obeys Zipf's law closely, reflecting its social context; iii) we show that even though ChiSCor is relatively small, the corpus is rich enough to train informative lemma vectors that allow us to analyse children's language use. We end with a reflection on the value of narrative datasets in computational linguistics.
163.
van Duijn M; van Dijk B; Kouwenhoven T; de Valk W; Spruit M; vanderPutten P
Theory of Mind in Large Language Models: Examining Performance of 11 State-of-the-Art models vs. Children Aged 7-10 on Advanced Tests Proceedings Article
In: Jiang, Jing; Reitter, David; Deng, Shumin (Ed.): Proceedings of the 27th Conference on Computational Natural Language Learning (CoNLL), pp. 389–402, Association for Computational Linguistics, Singapore, 2023.
@inproceedings{van-duijn-etal-2023-theory,
title = {Theory of Mind in Large Language Models: Examining Performance of 11 State-of-the-Art models vs. Children Aged 7-10 on Advanced Tests},
author = {van Duijn, Max and
van Dijk, Bram and
Kouwenhoven, Tom and
de Valk, Werner and
Spruit, Marco and
vanderPutten, Peter},
editor = {Jiang, Jing and
Reitter, David and
Deng, Shumin},
url = {https://aclanthology.org/2023.conll-1.25},
doi = {10.18653/v1/2023.conll-1.25},
year = {2023},
date = {2023-12-01},
booktitle = {Proceedings of the 27th Conference on Computational Natural Language Learning (CoNLL)},
pages = {389–402},
publisher = {Association for Computational Linguistics},
address = {Singapore},
abstract = {To what degree should we ascribe cognitive capacities to Large Language Models (LLMs), such as the ability to reason about intentions and beliefs known as Theory of Mind (ToM)? Here we add to this emerging debate by (i) testing 11 base- and instruction-tuned LLMs on capabilities relevant to ToM beyond the dominant false-belief paradigm, including non-literal language usage and recursive intentionality; (ii) using newly rewritten versions of standardized tests to gauge LLMs' robustness; (iii) prompting and scoring for open besides closed questions; and (iv) benchmarking LLM performance against that of children aged 7-10 on the same tasks. We find that instruction-tuned LLMs from the GPT family outperform other models, and often also children. Base-LLMs are mostly unable to solve ToM tasks, even with specialized prompting. We suggest that the interlinked evolution and development of language and ToM may help explain what instruction-tuning adds: rewarding cooperative communication that takes into account interlocutor and context. We conclude by arguing for a nuanced perspective on ToM in LLMs.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
To what degree should we ascribe cognitive capacities to Large Language Models (LLMs), such as the ability to reason about intentions and beliefs known as Theory of Mind (ToM)? Here we add to this emerging debate by (i) testing 11 base- and instruction-tuned LLMs on capabilities relevant to ToM beyond the dominant false-belief paradigm, including non-literal language usage and recursive intentionality; (ii) using newly rewritten versions of standardized tests to gauge LLMs' robustness; (iii) prompting and scoring for open besides closed questions; and (iv) benchmarking LLM performance against that of children aged 7-10 on the same tasks. We find that instruction-tuned LLMs from the GPT family outperform other models, and often also children. Base-LLMs are mostly unable to solve ToM tasks, even with specialized prompting. We suggest that the interlinked evolution and development of language and ToM may help explain what instruction-tuning adds: rewarding cooperative communication that takes into account interlocutor and context. We conclude by arguing for a nuanced perspective on ToM in LLMs.
162.
van Dijk B; Kouwenhoven T; Spruit M; van Duijn M J
Large Language Models: The Need for Nuance in Current Debates and a Pragmatic Perspective on Understanding Proceedings Article
In: Bouamor, Houda; Pino, Juan; Bali, Kalika (Ed.): Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pp. 12641–12654, Association for Computational Linguistics, Singapore, 2023.
@inproceedings{van-dijk-etal-2023-large,
title = {Large Language Models: The Need for Nuance in Current Debates and a Pragmatic Perspective on Understanding},
author = {van Dijk, Bram and
Kouwenhoven, Tom and
Spruit, Marco and
van Duijn, Max Johannes},
editor = {Bouamor, Houda and
Pino, Juan and
Bali, Kalika},
url = {https://aclanthology.org/2023.emnlp-main.779},
doi = {10.18653/v1/2023.emnlp-main.779},
year = {2023},
date = {2023-12-01},
booktitle = {Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing},
pages = {12641–12654},
publisher = {Association for Computational Linguistics},
address = {Singapore},
abstract = {Current Large Language Models (LLMs) are unparalleled in their ability to generate grammatically correct, fluent text. LLMs are appearing rapidly, and debates on LLM capacities have taken off, but reflection is lagging behind. Thus, in this position paper, we first zoom in on the debate and critically assess three points recurring in critiques of LLM capacities: i) that LLMs only parrot statistical patterns in the training data; ii) that LLMs master formal but not functional language competence; and iii) that language learning in LLMs cannot inform human language learning. Drawing on empirical and theoretical arguments, we show that these points need more nuance. Second, we outline a pragmatic perspective on the issue of `real' understanding and intentionality in LLMs. Understanding and intentionality pertain to unobservable mental states we attribute to other humans because they have pragmatic value: they allow us to abstract away from complex underlying mechanics and predict behaviour effectively. We reflect on the circumstances under which it would make sense for humans to similarly attribute mental states to LLMs, thereby outlining a pragmatic philosophical context for LLMs as an increasingly prominent technology in society.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Current Large Language Models (LLMs) are unparalleled in their ability to generate grammatically correct, fluent text. LLMs are appearing rapidly, and debates on LLM capacities have taken off, but reflection is lagging behind. Thus, in this position paper, we first zoom in on the debate and critically assess three points recurring in critiques of LLM capacities: i) that LLMs only parrot statistical patterns in the training data; ii) that LLMs master formal but not functional language competence; and iii) that language learning in LLMs cannot inform human language learning. Drawing on empirical and theoretical arguments, we show that these points need more nuance. Second, we outline a pragmatic perspective on the issue of `real' understanding and intentionality in LLMs. Understanding and intentionality pertain to unobservable mental states we attribute to other humans because they have pragmatic value: they allow us to abstract away from complex underlying mechanics and predict behaviour effectively. We reflect on the circumstances under which it would make sense for humans to similarly attribute mental states to LLMs, thereby outlining a pragmatic philosophical context for LLMs as an increasingly prominent technology in society.
161.
Heuss M; Cohen D; Mansoury M; de Rijke M; Eickhoff C
Predictive Uncertainty-based Bias Mitigation in Ranking Proceedings Article
In: CIKM 2023: 32nd ACM International Conference on Information and Knowledge Management, pp. 762–772, ACM, 2023.
@inproceedings{heuss-2023-predictive,
title = {Predictive Uncertainty-based Bias Mitigation in Ranking},
author = {Heuss, Maria and Cohen, Daniel and Mansoury, Masoud and de Rijke, Maarten and Eickhoff, Carsten},
url = {https://arxiv.org/abs/2309.09833},
year = {2023},
date = {2023-10-01},
booktitle = {CIKM 2023: 32nd ACM International Conference on Information and Knowledge Management},
pages = {762–772},
publisher = {ACM},
abstract = {Societal biases that are contained in retrieved documents have received increased interest. Such biases, which are often prevalent in the training data and learned by the model, can cause societal harms, by misrepresenting certain groups, and by enforcing stereotypes. Mitigating such biases demands algorithms that balance the trade-off between maximized utility for the user with fairness objectives, which incentivize unbiased rankings. Prior work on bias mitigation often assumes that ranking scores, which correspond to the utility that a document holds for a user, can be accurately determined. In reality, there is always a degree of uncertainty in the estimate of expected document utility. This uncertainty can be approximated by viewing ranking models through a Bayesian perspective, where the standard deterministic score becomes a distribution. In this work, we investigate whether uncertainty estimates can be used to decrease the amount of bias in the ranked results, while minimizing loss in measured utility. We introduce a simple method that uses the uncertainty of the ranking scores for an uncertainty-aware, post hoc approach to bias mitigation. We compare our proposed method with existing baselines for bias mitigation with respect to the utility-fairness trade-off, the controllability of methods, and computational costs. We show that an uncertainty-based approach can provide an intuitive and flexible trade-off that outperforms all baselines without additional training requirements, allowing for the post hoc use of this approach on top of arbitrary retrieval models.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Societal biases that are contained in retrieved documents have received increased interest. Such biases, which are often prevalent in the training data and learned by the model, can cause societal harms, by misrepresenting certain groups, and by enforcing stereotypes. Mitigating such biases demands algorithms that balance the trade-off between maximized utility for the user with fairness objectives, which incentivize unbiased rankings. Prior work on bias mitigation often assumes that ranking scores, which correspond to the utility that a document holds for a user, can be accurately determined. In reality, there is always a degree of uncertainty in the estimate of expected document utility. This uncertainty can be approximated by viewing ranking models through a Bayesian perspective, where the standard deterministic score becomes a distribution. In this work, we investigate whether uncertainty estimates can be used to decrease the amount of bias in the ranked results, while minimizing loss in measured utility. We introduce a simple method that uses the uncertainty of the ranking scores for an uncertainty-aware, post hoc approach to bias mitigation. We compare our proposed method with existing baselines for bias mitigation with respect to the utility-fairness trade-off, the controllability of methods, and computational costs. We show that an uncertainty-based approach can provide an intuitive and flexible trade-off that outperforms all baselines without additional training requirements, allowing for the post hoc use of this approach on top of arbitrary retrieval models.
160.
van Leeuwen L; Verheij B; Verbrugge R; Renooij S
Using Agent-Based Simulations to Evaluate Bayesian Networks for Criminal Scenarios. Proceedings Article
In: The Nineteenth International Conference on Artificial Intelligence and Law (ICAIL 2023). Proceedings of the Conference, ACM, New York, NY, USA, Braga, Portugal, 2023.
@inproceedings{vanLeeuwen2023,
title = {Using Agent-Based Simulations to Evaluate Bayesian Networks for Criminal Scenarios. },
author = {Ludi van Leeuwen and Bart Verheij and Rineke Verbrugge and Silja Renooij},
doi = {https://doi.org/10.1145/3594536.3595125},
year = {2023},
date = {2023-06-01},
booktitle = {The Nineteenth International Conference on Artificial Intelligence and Law (ICAIL 2023). Proceedings of the Conference},
publisher = {ACM, New York, NY, USA},
address = {Braga, Portugal},
abstract = {Scenario-based Bayesian networks (BNs) have been proposed as a tool for the rational handling of evidence. The proper evaluation of existing methods requires access to a ground truth that can be used to test the quality and usefulness of a BN model of a crime. However, that would require a full probability distribution over all relevant variables used in the model, which is in practice not available. In this paper, we use an agent-based simulation as a proxy for the ground truth for the evaluation of BN models as tools for the rational handling of evidence. We use fictional crime scenarios as a background. First, we design manually constructed BNs using existing design methods in order to model example crime scenarios. Second, we build an agent-based simulation covering the scenarios of criminal and non-criminal behavior. Third, we algorithmically determine BNs using statistics collected experimentally from the agent-based simulation that represents the ground truth. Finally, we compare the manual, scenario-based BNs to the algorithmic BNs by comparing the posterior probability distribution over outcomes of the network to the ground-truth frequency distribution over those outcomes in the simulation, across all evidence valuations. We find that both manual BNs and algorithmic BNs perform similarly well: they are good reflections of the ground truth in most of the evidence valuations. Using ABMs as a ground truth can be a tool to investigate Bayesian Networks and their design methods, especially under circumstances that are implausible in real-life criminal cases, such as full probabilistic information.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Scenario-based Bayesian networks (BNs) have been proposed as a tool for the rational handling of evidence. The proper evaluation of existing methods requires access to a ground truth that can be used to test the quality and usefulness of a BN model of a crime. However, that would require a full probability distribution over all relevant variables used in the model, which is in practice not available. In this paper, we use an agent-based simulation as a proxy for the ground truth for the evaluation of BN models as tools for the rational handling of evidence. We use fictional crime scenarios as a background. First, we design manually constructed BNs using existing design methods in order to model example crime scenarios. Second, we build an agent-based simulation covering the scenarios of criminal and non-criminal behavior. Third, we algorithmically determine BNs using statistics collected experimentally from the agent-based simulation that represents the ground truth. Finally, we compare the manual, scenario-based BNs to the algorithmic BNs by comparing the posterior probability distribution over outcomes of the network to the ground-truth frequency distribution over those outcomes in the simulation, across all evidence valuations. We find that both manual BNs and algorithmic BNs perform similarly well: they are good reflections of the ground truth in most of the evidence valuations. Using ABMs as a ground truth can be a tool to investigate Bayesian Networks and their design methods, especially under circumstances that are implausible in real-life criminal cases, such as full probabilistic information.
159.
Mehrotra S; Jorge C C; Jonker C M; Tielman M L
Building Appropriate Trust in AI: The Significance of Integrity-Centered Explanations Proceedings Article
In: HHAI 2023: Augmenting Human Intellect, pp. 436 - 439, IOS Press Ebooks, 2023.
@inproceedings{siddharth2023a,
title = {Building Appropriate Trust in AI: The Significance of Integrity-Centered Explanations},
author = {Mehrotra, Siddharth and Jorge, Carolina Centeio and Jonker, Catholijn M. and Tielman, Myrthe L.},
doi = {10.3233/FAIA230121},
year = {2023},
date = {2023-06-01},
booktitle = {HHAI 2023: Augmenting Human Intellect},
volume = {368},
pages = {436 - 439},
publisher = {IOS Press Ebooks},
series = {Frontiers in Artificial Intelligence and Applications},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
158.
Ligthart M E U; Neerincx M A; Hindriks K V
It Takes Two: Using Co-Creation to Facilitate Child-Robot Co-Regulation Journal Article
In: Transactions on Human-Robot Interaction, 2023.
@article{10.1145/3593812,
title = {It Takes Two: Using Co-Creation to Facilitate Child-Robot Co-Regulation},
author = {Ligthart, Mike E.U. and Neerincx, Mark A. and Hindriks, Koen V.},
url = {doi.org/10.1145/3593812},
doi = {10.1145/3593812},
year = {2023},
date = {2023-05-01},
journal = {Transactions on Human-Robot Interaction},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
abstract = {While interacting with a social robot, children have a need to express themselves and have their expressions acknowledged by the robot. A need that is often unaddressed by the robot, due to its limitations in understanding the expressions of children. To keep the child-robot interaction manageable the robot takes control, undermining children’s ability to co-regulate the interaction. Co-regulation is important for having a fulfilling social interaction. We developed a co-creation activity that aims to facilitate more co-regulation. Children are enabled to create sound effects, gestures, and light animations for the robot to use during their conversation. A crucial additional feature is that children are able to coordinate their involvement of the co-creation process. Results from a user study (N = 59 school children, 7-11 y.o.) showed that the co-creation activity successfully facilitated co-regulation by improving children’s agency. It also positively affected the acceptance of the robot. We furthermore identified five distinct profiles detailing the different needs and motivations children have for the level of involvement they chose during the co-creation process.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
While interacting with a social robot, children have a need to express themselves and have their expressions acknowledged by the robot. A need that is often unaddressed by the robot, due to its limitations in understanding the expressions of children. To keep the child-robot interaction manageable the robot takes control, undermining children’s ability to co-regulate the interaction. Co-regulation is important for having a fulfilling social interaction. We developed a co-creation activity that aims to facilitate more co-regulation. Children are enabled to create sound effects, gestures, and light animations for the robot to use during their conversation. A crucial additional feature is that children are able to coordinate their involvement of the co-creation process. Results from a user study (N = 59 school children, 7-11 y.o.) showed that the co-creation activity successfully facilitated co-regulation by improving children’s agency. It also positively affected the acceptance of the robot. We furthermore identified five distinct profiles detailing the different needs and motivations children have for the level of involvement they chose during the co-creation process.
157.
Orzan N; Acar E; Grossi D; Radulescu R
Emergent Cooperation and Deception in Public Good Games Conference
2023, (2023 Adaptive and Learning Agents Workshop at AAMAS, ALA 2023 ; Conference date: 29-05-2023 Through 30-05-2023).
@conference{42fc313afdcf4ea0a011c0fdb462ef1a,
title = {Emergent Cooperation and Deception in Public Good Games},
author = {Nicole Orzan and Erman Acar and Davide Grossi and Roxana Radulescu},
url = {https://alaworkshop2023.github.io},
year = {2023},
date = {2023-05-01},
abstract = {Communication is a widely used mechanism to promote cooperation in multi-agent systems. In the field of emergent communication agents are usually trained on a particular type of environment: cooperative, competitive, or mixed-motive. Motivated by the idea that real-world settings are characterised by incomplete information and that humans face daily interactions under a wide spectrum of incentives, we hypothesise that emergent communication could be simultaneously exploited in the totality of these scenarios. In this work we pursue this line of research by focusing on social dilemmas, and develop an extended version of the Public Goods Game which allows us to train independent reinforcement learning agents simultaneously on different scenarios where incentives are aligned (or misaligned) to various extents. Additionally, we introduce uncertainty regarding the alignment of incentives, and we equip agents with the ability to learn a communication policy, to study the potential of emergent communication for overcoming uncertainty. We show that in settings where all agents have the same level of uncertainty, communication can help improve the cooperation level of the system, while, when uncertainty is asymmetric, certain agents learn to use communication to deceive and exploit their uncertain peers.},
note = {2023 Adaptive and Learning Agents Workshop at AAMAS, ALA 2023 ; Conference date: 29-05-2023 Through 30-05-2023},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Communication is a widely used mechanism to promote cooperation in multi-agent systems. In the field of emergent communication agents are usually trained on a particular type of environment: cooperative, competitive, or mixed-motive. Motivated by the idea that real-world settings are characterised by incomplete information and that humans face daily interactions under a wide spectrum of incentives, we hypothesise that emergent communication could be simultaneously exploited in the totality of these scenarios. In this work we pursue this line of research by focusing on social dilemmas, and develop an extended version of the Public Goods Game which allows us to train independent reinforcement learning agents simultaneously on different scenarios where incentives are aligned (or misaligned) to various extents. Additionally, we introduce uncertainty regarding the alignment of incentives, and we equip agents with the ability to learn a communication policy, to study the potential of emergent communication for overcoming uncertainty. We show that in settings where all agents have the same level of uncertainty, communication can help improve the cooperation level of the system, while, when uncertainty is asymmetric, certain agents learn to use communication to deceive and exploit their uncertain peers.
156.
Kim T; Cochez M; François-Lavet V; Neerincx M; Vossen P
A Machine with Short-Term, Episodic, and Semantic Memory Systems Proceedings Article
In: Proceedings of the AAAI Conference on Artificial Intelligence, 2023.
@inproceedings{Taewoo-AAAI-2023,
title = {A Machine with Short-Term, Episodic, and Semantic Memory Systems},
author = {Kim, Taewoon and Cochez, Michael and François-Lavet, Vincent and Neerincx, Mark and Vossen, Piek},
url = {https://arxiv.org/abs/2212.02098},
doi = {10.48550/ARXIV.2212.02098},
year = {2023},
date = {2023-02-01},
booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
abstract = {Inspired by the cognitive science theory of the explicit human memory systems, we have modeled an agent with short-term, episodic, and semantic memory systems, each of which is modeled with a knowledge graph. To evaluate this system and analyze the behavior of this agent, we designed and released our own reinforcement learning agent environment, "the Room", where an agent has to learn how to encode, store, and retrieve memories to maximize its return by answering questions. We show that our deep Q-learning based agent successfully learns whether a short-term memory should be forgotten, or rather be stored in the episodic or semantic memory systems. Our experiments indicate that an agent with human-like memory systems can outperform an agent without this memory structure in the environment.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Inspired by the cognitive science theory of the explicit human memory systems, we have modeled an agent with short-term, episodic, and semantic memory systems, each of which is modeled with a knowledge graph. To evaluate this system and analyze the behavior of this agent, we designed and released our own reinforcement learning agent environment, "the Room", where an agent has to learn how to encode, store, and retrieve memories to maximize its return by answering questions. We show that our deep Q-learning based agent successfully learns whether a short-term memory should be forgotten, or rather be stored in the episodic or semantic memory systems. Our experiments indicate that an agent with human-like memory systems can outperform an agent without this memory structure in the environment.
155.
Sauter A W M; Acar E; Francois-Lavet V
A Meta-Reinforcement Learning Algorithm for Causal Discovery Proceedings Article
In: 2nd Conference on Causal Learning and Reasoning, 2023.
@inproceedings{sauter2023a,
title = {A Meta-Reinforcement Learning Algorithm for Causal Discovery},
author = {Andreas W.M. Sauter and Erman Acar and Vincent Francois-Lavet},
url = {https://openreview.net/forum?id=p6NnDqJM_jL},
year = {2023},
date = {2023-01-01},
booktitle = {2nd Conference on Causal Learning and Reasoning},
abstract = {Uncovering the underlying causal structure of a phenomenon, domain or environment is of great scientific interest, not least because of the inferences that can be derived from such structures. Unfortunately though, given an environment, identifying its causal structure poses significant challenges. Amongst those are the need for costly interventions and the size of the space of possible structures that has to be searched. In this work, we propose a meta-reinforcement learning setup that addresses these challenges by learning a causal discovery algorithm, called Meta-Causal Discovery, or MCD. We model this algorithm as a policy that is trained on a set of environments with known causal structures to perform budgeted interventions. Simultaneously, the policy learns to maintain an estimate of the environment's causal structure. The learned policy can then be used as a causal discovery algorithm to estimate the structure of environments in a matter of milliseconds. At test time, our algorithm performs well even in environments that induce previously unseen causal structures. We empirically show that MCD estimates good graphs compared to SOTA approaches on toy environments and thus constitutes a proof-of-concept of learning causal discovery algorithms.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Uncovering the underlying causal structure of a phenomenon, domain or environment is of great scientific interest, not least because of the inferences that can be derived from such structures. Unfortunately though, given an environment, identifying its causal structure poses significant challenges. Amongst those are the need for costly interventions and the size of the space of possible structures that has to be searched. In this work, we propose a meta-reinforcement learning setup that addresses these challenges by learning a causal discovery algorithm, called Meta-Causal Discovery, or MCD. We model this algorithm as a policy that is trained on a set of environments with known causal structures to perform budgeted interventions. Simultaneously, the policy learns to maintain an estimate of the environment's causal structure. The learned policy can then be used as a causal discovery algorithm to estimate the structure of environments in a matter of milliseconds. At test time, our algorithm performs well even in environments that induce previously unseen causal structures. We empirically show that MCD estimates good graphs compared to SOTA approaches on toy environments and thus constitutes a proof-of-concept of learning causal discovery algorithms.
154.
Han S; Dastani M; Wang S
Model-based Sparse Communication in Multi-agent Reinforcement Learning Proceedings Article
In: Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems, pp. 439–447, 2023.
@inproceedings{han2023model,
title = {Model-based Sparse Communication in Multi-agent Reinforcement Learning},
author = {Han, Shuai and Dastani, Mehdi and Wang, Shihan},
url = {https://www.southampton.ac.uk/~eg/AAMAS2023/pdfs/p439.pdf},
year = {2023},
date = {2023-01-01},
booktitle = {Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems},
pages = {439–447},
abstract = {Learning to communicate efficiently is central to multi-agent reinforcement learning (MARL). Existing methods often require agents to exchange messages intensively, which abuses communication channels and leads to high communication overhead. Only a few methods target on learning sparse communication, but they allow limited information to be shared, which affects the efficiency of policy learning. In this work, we propose model-based communication (MBC), a learning framework with a decentralized communication scheduling process. The MBC framework enables multiple agents to make decisions with sparse communication. In particular, the MBC framework introduces a model-based message estimator to estimate the up-to-date global messages using past local data. A decentralized message scheduling mechanism is also proposed to determine whether a message shall be sent based on the estimation. We evaluated our method in a variety of mixed cooperative-competitive environments. The experiment results show that the MBC method shows better performance and lower channel overhead than the state-of-art baselines.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Learning to communicate efficiently is central to multi-agent reinforcement learning (MARL). Existing methods often require agents to exchange messages intensively, which abuses communication channels and leads to high communication overhead. Only a few methods target on learning sparse communication, but they allow limited information to be shared, which affects the efficiency of policy learning. In this work, we propose model-based communication (MBC), a learning framework with a decentralized communication scheduling process. The MBC framework enables multiple agents to make decisions with sparse communication. In particular, the MBC framework introduces a model-based message estimator to estimate the up-to-date global messages using past local data. A decentralized message scheduling mechanism is also proposed to determine whether a message shall be sent based on the estimation. We evaluated our method in a variety of mixed cooperative-competitive environments. The experiment results show that the MBC method shows better performance and lower channel overhead than the state-of-art baselines.
153.
Ulfert A; Georganta E; Jorge C C; Mehrotra S; Tielman M
Shaping a multidisciplinary understanding of team trust in human-AI teams: a theoretical framework Journal Article
In: European Journal of Work and Organizational Psychology, pp. 1–14, 2023.
@article{doi:10.1080/1359432X.2023.2200172,
title = {Shaping a multidisciplinary understanding of team trust in human-AI teams: a theoretical framework},
author = {Anna-Sophie Ulfert and Eleni Georganta and Carolina Centeio Jorge and Siddharth Mehrotra and Myrthe Tielman},
url = {https://www.tandfonline.com/doi/full/10.1080/1359432X.2023.2200172},
doi = {10.1080/1359432X.2023.2200172},
year = {2023},
date = {2023-01-01},
journal = {European Journal of Work and Organizational Psychology},
pages = {1–14},
publisher = {Routledge},
abstract = {Intelligent systems are increasingly entering the workplace, gradually moving away from technologies supporting work processes to artificially intelligent (AI) agents becoming team members. Therefore, a deep understanding of effective human-AI collaboration within the team context is required. Both psychology and computer science literature emphasize the importance of trust when humans interact either with human team members or AI agents. However, empirical work and theoretical models that combine these research fields and define team trust in human-AI teams are scarce. Furthermore, they often lack to integrate central aspects, such as the multilevel nature of team trust and the role of AI agents as team members. Building on an integration of current literature on trust in human-AI teaming across different research fields, we propose a multidisciplinary framework of team trust in human-AI teams. The framework highlights different trust relationships that exist within human-AI teams and acknowledges the multilevel nature of team trust. We discuss the framework’s potential for human-AI teaming research and for the design and implementation of trustworthy AI team members.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Intelligent systems are increasingly entering the workplace, gradually moving away from technologies supporting work processes to artificially intelligent (AI) agents becoming team members. Therefore, a deep understanding of effective human-AI collaboration within the team context is required. Both psychology and computer science literature emphasize the importance of trust when humans interact either with human team members or AI agents. However, empirical work and theoretical models that combine these research fields and define team trust in human-AI teams are scarce. Furthermore, they often lack to integrate central aspects, such as the multilevel nature of team trust and the role of AI agents as team members. Building on an integration of current literature on trust in human-AI teaming across different research fields, we propose a multidisciplinary framework of team trust in human-AI teams. The framework highlights different trust relationships that exist within human-AI teams and acknowledges the multilevel nature of team trust. We discuss the framework’s potential for human-AI teaming research and for the design and implementation of trustworthy AI team members.
152.
Onnes A; Dastani M; Renooij S
Bayesian Network Conflict Detection for Normative Monitoring of Black-Box Systems Proceedings Article
In: Proceedings of the Thirty-Sixth FLAIRS Conference, Florida Online Journals, 2023.
@inproceedings{onnes2023,
title = {Bayesian Network Conflict Detection for Normative Monitoring of Black-Box Systems},
author = {Onnes, Annet and Dastani, Mehdi and Renooij, Silja},
url = {https://journals.flvc.org/FLAIRS/article/view/133240/137859},
year = {2023},
date = {2023-01-01},
booktitle = {Proceedings of the Thirty-Sixth FLAIRS Conference},
volume = {36},
publisher = {Florida Online Journals},
abstract = {Bayesian networks are interpretable probabilistic models that can be constructed from both data and domain knowledge. They are applied in various domains and for different tasks, including that of anomaly detection, for which an easy to compute measure of data conflict exists. In this paper we consider the use of Bayesian networks to monitor input-output pairs of a black-box AI system, to establish whether the output is acceptable in the current context in which the AI system operates. A Bayesian network-based prescriptive, or normative, model is assumed that includes context variables relevant for deciding what is or is not acceptable. We analyse and adjust the conflict measure to make it applicable to our new type of monitoring setting.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Bayesian networks are interpretable probabilistic models that can be constructed from both data and domain knowledge. They are applied in various domains and for different tasks, including that of anomaly detection, for which an easy to compute measure of data conflict exists. In this paper we consider the use of Bayesian networks to monitor input-output pairs of a black-box AI system, to establish whether the output is acceptable in the current context in which the AI system operates. A Bayesian network-based prescriptive, or normative, model is assumed that includes context variables relevant for deciding what is or is not acceptable. We analyse and adjust the conflict measure to make it applicable to our new type of monitoring setting.