Auteur : Dagenais, Mylène

Staffan LARSSON - 8 avril 2021

Titre: Meaning as coordinated and composable classifiers

Résumé:

How are meanings of utterances related to the world and our perception of it? What is meaning, and how is it created? How do word meanings contribute to utterance meaning? We are working towards a formal semantics that aims to provide answers to these and related questions, starting from situated interaction between agents. The meanings of many expressions can be modeled as classifiers of real-world information. Expressions can be single words, or phrases and sentences whose meanings are composed from the meanings of their constituents. By interacting, agents coordinate on meanings by training classifiers. To make formally explicit the notions of coordination, compositionality and classification, and to relate these notions to each other, we use TTR (a type theory with records).

Références:

- Larsson, S. (2015). Formal semantics for perceptual classification. Journal of logic and computation, 25(2), 335-369. https://academic.oup.com/logcom/article-abstract/25/2/335/954129

- Larsson, S. (2018). Grounding as a side‐effect of grounding. Topics in cognitive science, 10(2), 389-408. https://onlinelibrary.wiley.com/doi/full/10.1111/tops.12317

- Larsson, S. (2020). Discrete and Probabilistic Classifier-based Semantics. In Proceedings of the Probability and Meaning Conference (PaM 2020) (pp. 62-68). https://www.aclweb.org/anthology/2020.pam-1.8.pdf

Bio:

Staffan Larsson (b. 1969) was educated at University of Gothenburg (1992-1996) and gained a PhD in Linguistics there (1997-2002). Since 2013, he is Professor of Computational Linguistics at the Department of Philosophy, Linguistics and Theory of Science at the University of Gothenburg. He is also a member of CLASP (Centre for Research on Linguistic Theory and Studies in Probability) and co-founder and Chief Science Officer of Talkamatic AB. His areas of interest include dialogue, dialogue systems, language and perception, pragmatics, formal semantics, semantic coordination, and philosophy of language.

Alberto TESTOLIN - 1er avril 2021

Titre: The challenge of modeling the acquisition of mathematical concepts.

Résumé:

Mathematics is one of the most impressive achievements of human cultural evolution. Despite we perceive it as being overly abstract, it is widely believed that mathematical skills are rooted into a phylogenetically ancient “number sense”, which allows us to approximately represent quantities. However, the relationship between number sense and the subsequent acquisition of symbolic mathematical concepts remains controversial. In this seminar I will discuss how recent advances in AI and deep learning research might allow to investigate how the acquisition of numerical concepts could be grounded into sensorimotor experiences. Success in this challenging enterprise would have immediate implications for cognitive science, but also far-reaching impact for educational practice and for the creation of the next generation of intelligent machines.

Références:

1) Zorzi, M., & Testolin, A. (2018). An emergentist perspective on the origin of number sense. Philosophical Transactions of the Royal Society B: Biological Sciences, 373(1740), 20170043. https://royalsocietypublishing.org/doi/full/10.1098/rstb.2017.0043

2) Overmann, K. A. (2018). Constructing a concept of number. Journal of Numerical Cognition. 4, 464–493. https://jnc.psychopen.eu/article/view/161/html

Bio:

Alberto Testolin received the M.Sc. degree in Computer Science and the Ph.D. degree in Psychological Sciences from the University of Padova, Italy, in 2011 and 2015, respectively. In 2019 he was Visiting Scholar at the Department of Psychology at Stanford University. He is currently Assistant Professor at the University of Padova, with a joint appointment at the Department of Information Engineering and the Department of General Psychology. He is broadly interested in artificial intelligence, machine learning and cognitive neuroscience. His main research interests are statistical learning theory, predictive coding, sensory perception, cognitive modeling and applications of deep learning to signal processing and optimization. He is an active member of the IEEE Task Force on Deep Learning.

Jun TANI - 31 mars 2021

Titre: Exploring robotic minds using predictive coding and active inference frameworks

Résumé:

The focus of my research has been to investigate how cognitive agents can acquire structural representation via iterative interaction with the world, exercising agency and learning from resultant perceptual experience. For this purpose, my group has investigated various models analogous to predictive coding and active inference frameworks. For the past two decades, we have applied these frameworks to develop cognitive constructs for robots. My talk attempts to clarify underlying cognitive and mind mechanisms for compositionality, social cognition, and consciousness from analysis of emergent phenomena observed in these robotics experiments.

Références:

(1) Tani, J. (2016). “Exploring Robotic Minds: Actions, Symbols, and Consciousness as Self-Organizing Dynamic Phenomena.”, Oxford University Press. link

(2) Tani, J., & White, J. (2020). Cognitive neurorobotics and self in the shared world, a focused review of ongoing research. Adaptive Behavior, 1059712320962158.

Bio:

Jun Tani received the D.Eng. degree from Sophia University, Tokyo in 1995. He started his research career with Sony Computer Science Lab. in 1993. He became a Team Leader of the Laboratory for Behavior and Dynamic Cognition, RIKEN Brain Science Institute, Saitama, Japan in 2001. He became a Full Professor with the Electrical Engineering Department, Korea Advanced Institute of Science and Technology, Daejeon, South Korea in 2012. He is currently a Full Professor with the Okinawa Institute of Science and Technology, Okinawa, Japan. His current research interests include cognitive neuroscience, developmental psychology, phenomenology, complex adaptive systems, and robotics.

Gary LUPYAN - 25 mars 2021

Titre: From Whorf to Telepathy: How words structure and align our concepts

Résumé:

That people are able to communicate on a wide range of topics with reasonable success is often taken as evidence that we have a largely overlapping conceptual repertoire. But where do our concepts come from and how similar are they, really? On one widespread view, humans are born with a core-knowledge system and a set of conceptual categories onto which words map. Alternatively, many of our concepts — including some that seem very basic — may derive from our experience with and use of language. On this view, language plays a key role in both constructing and aligning our conceptual spaces. I will argue in favor of the second view, present evidence for the causal role of language in categorization and reasoning, and describe what consequences this position has for the theoretical possibility of telepathy.

Bio:

Gary Lupyan is a professor of psychology at University of Wisconsin-Madison. He obtained his doctorate in 2007 at Carnegie Mellon with Jay McClelland, followed by postdocs in cognitive (neuro)science at Cornell University and University of Pennsylvania. At the center of his research interests is the question of whether and how our cognition and perception is augmented by language. What does language *do* for us? Other major research interests have spanned top-down effects in perception, the evolution of language, iconicity, and causes of linguistic diversity (do languages adapt to different socio-demographic environments?).

Tadahiro TANIGUCHI – 18 mars 2021

Titre: Symbol Emergence in Robotics: Probabilistic Generative Models for Real-world Multimodal Language Acquisition and Understanding

Résumé:

Symbol emergence in robotics aims to develop a robot that can adapt to the real-world environment, human linguistic communications, and acquire language from sensorimotor information alone, i.e., in an unsupervised manner. This line of studies is essential not only for creating a robot that can collaborate with people through human-robot interactions but also for understanding human cognitive development. This invited lecture introduces the recent development of integrative probabilistic generative models for language learning, e.g., spatial concept formation with simultaneous localization and mapping, and vision of symbol emergence in robotics. I will also introduce challenges related to the integration of probabilistic generative models and deep learning for language learning by robots.

Bio:

Tadahiro Taniguchi received the ME and Ph.D. degrees from Kyoto University, in 2003 and 2006, respectively. From April 2008 to March 2010, he was an Assistant Professor at the Department of Human and Computer Intelligence, Ritsumeikan University. From April 2010 to March 2017, he was an Associate Professor at the same department. From September 2015 to September 2016, he is a Visiting Associate Professor at the Department of Electrical and Electronic Engineering, Imperial College London. From April 2017, he has been a Professor at the Department of Information Science and Engineering, Ritsumeikan University. From April 2017, he has been a visiting general chief scientist, Technology Division, Panasonic, as well. He has been engaged in research on AI, symbol emergence in robotics, machine learning, and cognitive science.

Oliver ROESLER – 4 mars 2021

Titre : Combining unsupervised and supervised grounding approaches

Résumé:

There exist a variety of grounding approaches that either utilize supervised or unsupervised learning techniques to ground words through corresponding percepts. Supervised approaches are usually sample efficient but depend on the availability and trustworthiness of a tutor, while unsupervised approaches avoid this dependency, yet, they are less sample efficient and often also less accurate. So far, only limited work has been done to combine both approaches. In this talk, I will present recent work on combining cross-situational learning and interactive learning approaches to enable artificial language learners to benefit from the support and feedback of another agent, e.g., a human, without depending on it.

Références:

Roesler, O. (2020). Unsupervised Online Grounding of Natural Language during Human-Robot Interactions. arXiv preprint arXiv:2007.04304.

Roesler, O. (2020). Enhancing Unsupervised Natural Language Grounding through Explicit Teaching. In Proceedings of The 3rd UK-RAS Conference.

Bio:

Oliver Roesler, Brain Embodiment Lab, University of Reading; AI Lab, Vrije Universiteit Brussel; and Modality.AI, Inc. focuses on the development of mechanisms to enable natural, adaptive, and open-ended human-agent interactions through an interdisciplinary approach that combines work in knowledge representation and reasoning, language grounding, reinforcement learning based action learning as well as multimodal prediction of human affective and mental states.

Tony BELPAEME – 25 février 2021

Titre : The meaning of it all: Human-Robot Interaction

Résumé:

The representations used by the artificial intelligence powering robots are notoriously devoid of meaning: no computer or robot truly understands what it is doing or expressing, which is at the heart of the symbol grounding problem. Nevertheless, advances in AI seem to produce systems that on the surface appear to understand natural language, the most spectacular of which are the recent language models, such as GPT3. However, even though they are surprisingly effective at some tasks, they are merely statistical models lacking grounded semantics, something which explains their often surprising and erratic responses. In this talk I will explore if adding sensors and actuators to AI, i.e. building a robot, will endow AI with meaning. But while the robot does ground symbols, it is unlikely that these will be sufficiently aligned with ours to be useful for interaction. Instead, I argue that it is interaction itself between people and AI that will be needed for semantics to become not only grounded but also shared. We will propose how the interaction with robots can be used to allow AI to build semantic representation that are sufficiently aligned to allow robots and humans to share meaning.

Bio:

Tony Belpaeme is Professor at Ghent University and Professor of Cognitive Systems and Robotics at Plymouth University. He is a member of IDLab – imec at Ghent and is associated with the Centre for Robotics and Neural Systems at Plymouth. His research interests include social systems, cognitive robotics, and artificial intelligence in general.

Christophe SABOURIN – 18 février 2021

Titre : Systèmes cognitifs artificiels : du concept au développement de comportements intelligents en robotique autonome

Résumé:

Si le 20ème siècle a vu l’essor de la robotique industrielle principalement dans l’industrie manufacturière, le 21ème siècle sera certainement la période qui verra émerger la robotique de service ou les hommes et les robots devront apprendre à cohabiter et interagir dans un environnement partagé. Ces systèmes robotisés devront donc faire preuve d’une très grande souplesse d’adaptation en développant, au fil de leurs expériences de nouvelles capacités cognitives leur permettant d’apprendre progressivement à cohabiter et à collaborer avec les êtres humains. Les travaux présentés dans cet exposé s’appuient sur le principe de la robotique cognitive et plus particulièrement sur le paradigme de la cognition incarnée. La cognition associée au robot est donc le résultat d’un processus de développement où le robot devient progressivement plus habile et acquiert des connaissances lui permettant d’interpréter le monde qui l’entoure.

Références:

Sabourin, C., 2016. Systèmes cognitifs artificiels : du concept au développement de comportements intelligents en robotique autonome (Habilitation à diriger des recherches). Université Paris Est Créteil.

Ramík, D.M., Madani, K., Sabourin, C., 2013. From visual patterns to semantic description: A cognitive approach using artificial curiosity as the foundation. Pattern Recognition Letters 34, 1577–1588. https://doi.org/10.1016/j.patrec.2013.05.014

Bio:

Christophe Sabourin est enseignant chercheur en robotique et en intelligence artificielle. Il a obtenu son doctorat en 2004 à l'Université d'Orléans puis son Habilitation à diriger des Recherches en 2016 à l'Université PARIS-EST. Depuis 2005, il est Maître de Conférence à l'Université PARIS-EST Créteil (UPEC). Ses recherches portent sur le développement de systèmes cognitifs artificiels pour la robotique

André TRICOT – 11 février 2021

Titre : Mémoire humaine / mémoires artificielles

Résumé:

Depuis les gravures pariétales il y a 35 000 ans, les humains ont confié une partie de leurs souvenirs et de leurs connaissances à des mémoires externes. L’évolution de notre espèce nous a doté d’un système mnésique que nous accusons de défaillances majeures (oubli, distraction, blocage, etc.). A chaque innovation documentaire (tablettes d’argile, papyrus, imprimerie, photographie, web) nous avons augmenté la puissance de ces mémoires externes, déplorant aussitôt l’affaiblissement consécutif de notre mémoire naturelle. Pour développer des mémoires externes véritablement au service de notre mémoire naturelle il faudrait selon moi bien comprendre les fonctions des mémoires externes, et concevoir des accès et des organisations qui soulagent au lieu de surcharger notre mémoire.

Références:

Sahut, G., & Tricot, A. (2017). Social web and Wikipedia: an opportunity to rethink the links between sources' credibility, trust and authority. First Monday, 22, 6 November https://firstmonday.org/ojs/index.php/fm/article/view/7108/6555

Buckland, M. (2018). Document theory. ISKO Encyclopedia of Knowledge Organization. https://escholarship.org/content/qt64d1v86q/qt64d1v86q.pdf

Bio:

André Tricot est professeur de psychologie cognitive à l’Université Paul Valéry Montpellier 3 et chercheur au sein du laboratoire Epsylon (Dynamique des Capacités Humaines et des Conduites de Santé). Il s’intéresse aux relations entre les mémoires naturelles et artificielles. Il essaie de comprendre comment la conception d'une mémoire artificielle (un document) peut aider la mémoire naturelle au lieu de la surcharger. Les applications relèvent de l'ingénierie pédagogique, des interactions humain-machine, de l'ergonomie et de la sécurité des transports.