Auteur : Dagenais, Mylène

Séminaire DIC-ISC-CRIA – 22 septembre 2022 avec Mehdi KHAMASSI

Mehdi KHAMASSI – 22 septembre 2022

Titre : Active exploration in reinforcement learning: From neuroscience to robotics and vice versa

Résumé :

One of the key ingredients of learning in autonomous agents in volatile environments is the exploration-exploitation trade-off: finding the right balance between exploiting previously acquired knowledge and exploring alternatives, and adapting this balance on-the-fly when the environment changes. Throughout the presentation, I will use an illustrative example from the human–robot interaction (HRI) domain: Among relevant signals, non-verbal cues such as the human’s gaze can provide the robot with important information about the human’s current engagement in the task, and whether the robot should continue its current behavior or not. Various solutions have been proposed in the reinforcement learning literature, often inspired by developmental psychology (studying how human infants explore their surrounding world). Some mechanisms have neurobiological counterparts in the human brain: dynamic regulations of exploration rate as a function of volatility; information (uncertainty)-based solutions; and progress-based solutions. I will also illustrate existing bridges with Karl Friston’s active inference which he will later present in this seminar series.

Bio :

Mehdi Khamassi is a CNRS research director, Institute of Intelligent Systems and Robotics (ISIR), Sorbonne Université, Paris. His background is in Computer Science, Cognitive Sciences and Cognitive Neuroscience. He is co-director of studies of the CogMaster program at Ecole Normale Supérieure (PSL) / EHESS / University of Paris  and Editor of the several scientific journals, like Intellectica, Frontiers in Neurorobotics, Frontiers in Decision Neuroscience, ReScience X, and Neurons, Behavior, Data analysis and Theory. His main topics of research include decision-making and reinforcement learning in robots and humans, and the role of social and non-social rewards in learning.

References:

https://hal.archives-ouvertes.fr/hal-03415847/document

http://sites.isir.upmc.fr/www/files/2018ACLI4582.pdf

Séminaire DIC-ISC-CRIA – 15 septembre 2022 avec Jean-Pierre BRIOT

Titre : Music creation with deep learning techniques: Achievements and challenges

Résumé :

A growing application area for the current wave of deep learning (the return of artificial neural networks on steroids) is the generation of creative content, notably the case of music (and also images and text). The motivation is in using machine learning techniques to automatically learn musical styles from arbitrary musical corpora and then to generate musical samples from the estimated distribution, with some degree of control over the generation. This talk will survey some recent achievements in deep-learning-based music generation, using recent and dedicated generative architectures such as VAE, GAN and Transformer, analyzing principles, successes as well as challenges, including the limits of automated generation versus providing assistance to human musicians.

Bio :

Jean-Pierre BRIOT is a senior researcher (research director) in computer science at LIP6, joint computer science research lab of CNRS (Centre National de la Recherche Scientifique) and Sorbonne Université in Paris, France. He is also permanent visiting professor at PUC-Rio in Rio de Janeiro, Brazil. His general research interests are the design of intelligent adaptive and cooperative software, at the crossroads of artificial intelligence, distributed systems and software engineering, with various applications in the internet of things, decision support systems and computer music. His current interest is the use of AI techniques (notably deep learning-based) within music creation processes. He is the principal author of a recent reference book on deep learning techniques for music generation (Springer, 2020). https://link.springer.com/book/10.1007/978-3-319-70163-9

For more details (including access to publications): http://webia.lip6.fr/~briot/cv/

Briot, J. P. (2021). From artificial neural networks to deep learning for music generation: history, concepts and trends. Neural Computing and Applications, 33(1), 39-65.

https://hal.sorbonne-universite.fr/hal-02539189v3/file/nn4music-hal-v3.pdf

Briot, J. P. (2019). Apprentissage profond et génération de musique, Hors série Intelligence artificielle, Tangente – L’aventure mathématique, (68):30-37, September 2019.

https://webia.lip6.fr/~briot/cv/apgm-2019

Séminaire DIC-ISC-CRIA – 8 septembre 2022 avec Bernard J. BAARS

Bernard J. BAARS – 8 septembre 2022

Titre : Machine Consciousness Is Only a Metaphor

Résumé :

(V.F.) Les métaphores de la machine sont souvent utilisées en science, mais il ne faut pas les confondre avec la réalité. Toutes les cultures ont des métaphores concernant l’esprit. On y compte la caverne de Platon, le sens commun d’Aristote et le chariot d’Arjuna. Il ne faut pas confondre ces métaphores avec la conscience réelle. L’anthropologie ainsi que l’histoire humaine indiquent que les conflits humains sont souvent déclenchés par les mots déshumanisants. Attribuer la conscience à la machine parait innocent, mais ses implications ne le sont pas. Le cerveau conscient est une propriété biologique émergente, ayant des ramifications psychologiques et culturelles sans fin. L’étude empirique du cerveau conscient est en train de se redécouvrir. Ce champ de recherche a déjà été dérouté une fois par la spéculation indisciplinée. Ne répétons pas cette erreur.

(V.O.) Machine metaphors are often used in science but should not be confused with reality. All cultures have metaphors for mind. Examples are Plato’s Cave, Aristotle’s common sense and Arjuna’s chariot. These metaphors should not be confused with real consciousness; the implications of such a confusion can be dangerous. Anthropology and human history suggest that human conflicts tend to start with dehumanizing words. Machine consciousness sounds innocent, but its implications are not. The conscious brain is an emergent biological property, with endless psychological and cultural ripples. We are only beginning to rediscover the empirical study of the conscious brain in the sciences. Undisciplined speculation has already once undone this endeavor. Let’s not repeat that.

Bio :

Bernard J. Baars, Distinguished Senior Fellow at the Center for the Future Mind, Florida Atlantic University, is best known as the originator of the Global Workspace Theory (GWT) of conscious perception and cognition. After its initial development, GWT was applied to the cortico-thalamic system, a highly conserved mammalian structure, with analogs in other animal species. Other scientists have developed a “GWT family” of theories. Baars received the 2019 Hermann von Helmholtz Life Contribution Award from the International Neural Network Society, which “recognizes work in perception proven to be paradigm changing and long-lasting.”

Baars, B., Franklin, S., & Ramsøy, T. (2013). Global Workspace Dynamics: Cortical “Binding and Propagation” Enables Conscious Contents. Frontiers in Psychology, 4. https://doi.org/10.3389/fpsyg.2013.00200

https://www.frontiersin.org/articles/10.3389/fpsyg.2013.00200

Deco, G., Vidaurre, D. & Kringelbach, M.L. Revisiting the global workspace orchestrating the hierarchical organization of the human brain. Nat Hum Behav 5, 497–511 (2021). https://doi.org/10.1038/s41562-020-01003-6  

Baars, B. J., Geld, N., & Kozma, R. (2021). Global workspace theory (GWT) and prefrontal cortex: Recent developments. Frontiers in Psychology, 5163.

https://www.frontiersin.org/articles/10.3389/fpsyg.2021.749868/pdf

Le Moal, Michel. « Conclusion du colloque. » la lettre de l’Académie des sciences https://www.academie-sciences.fr/pdf/lettre/lettre30.pdf#page=20

Séminaire DIC-ISC-CRIA – 7 avril 2022 avec Arthur GLENGERG

Arthur GLENBERG – 7 avril 2022

Titre : The Centrality of Emotion to Language

Résumé:

Think about the content of language in venues other than academia (e.g., conversations among friends, novels, movies, musical lyrics, etc.): it is predominantly about emotions and feelings. Furthermore, the relation between emotion and language is not just in regard to content. I will review literature demonstrating the impact of emotion at multiple levels of language including phonology, prosody, syntax, word meaning, sentence comprehension, and discourse structure. The talk will end with a couple of suggestions for why there is this strong connection and what it implies. Because this is a new area for me, I don’t have a lot of my own work to cite. I have attached a couple of articles, but my advice to students is to skim them: They play a role in the talk but none is central. Finally, now that this is arranged, I will reveal my non-negotiable fee: a Université du Québec à Montréal t-shirt. For several years, I have been collecting t-shirts from places where I give talks. I think that I may now have the world’s greatest collection of university t-shirts. They make great souvenirs (in both the English and French etymological meanings), they are functional, and wearing one is a subtle way of showing off.

Bio:

Arthur GLENBERG is an emeritus professor in the Departments of Psychology at Arizona State University and the University of Wisconsin-Madison, and he has an appointment at INICO at the University of Salamanca in Spain. Glenberg conducts basic research in cognitive psychology and cognitive neuroscience. He has also developed a reading comprehension intervention for children in the early elementary grades based on principles of embodied cognition (Moved by Reading; https://www.movedbyreading.com/), and he has extended the intervention for English Language Learning children (EMBRACE). He is PI on an NSF-funded project directed at applying principles of embodied cognition to education. Glenberg has published a textbook (in its third edition), an edited volume, and over 140 peer-reviewed articles. His work has been cited more than 24,000 times, and his Google Scholar h-index is 63

Séminaire DIC-ISC-CRIA – 31 mars 2022 avec Pascale ZARATÉ

Pascale ZARATÉ – 31 mars 2022

Titre : From Decision Support Systems to Recommender Systems

Résumé:

Decision Support Systems (DSSs) emerged in the early 1970s. We can find in the literature several definitions of these systems and a large amount of work has been done. During all these years of research the architecture of these systems has evolved, and we assisted to the introduction of Artificial Intelligence components giving to these systems the name of Recommender systems. Recommender systems are now very used in daily life platforms, i.e., Amazon etc. The way how Recommender systems are developed and implemented, considering users’ preferences, will be presented in this talk.

Bio:

Pascale ZARATÉ is a Professor at Toulouse 1 Capitole University. She conducts her research at the IRIT laboratory (https://www.irit.fr/~Pascale.Zarate/ ). She holds a Ph.D. in Computer Sciences / Decision Support from the LAMSADE laboratory at the Paris Dauphine University, Paris (1991). Pascale Zaraté’s current research interests include Decision Support Systems, Group Decision Support Systems, Recommender systems She published several manuscripts: 3 books, edited 6 books, edited 18 special issues in several international journals, 11 proceedings of international conferences, 30 papers in international journals, 2 papers in national journals, 7 chapters in collective books, 52 papers in international conferences.

Références:

2021 Supporting multi-criteria decision-making across websites: the Logikós approach. Central European Journal of Operations Research (CEJOR), Springer Verlag (2021) 29(1), pp. 201-225, with Alejandro Fernandez, Juan Cruz Gardey, Gabriela Bosseti, DOI: 10.1007/s10100-020-00723-4

2019 A Combined Compromise Solution (CoCoSo) method for multi-criteria decision-making problems. Management Decision Emerald (2019) 57(9), pp 2501-2519, with Morteza Yazdani, Edmundas Kazimieras Zavadskas, Zenonas Turskis, DOI: 10.1108/MD-05-2017-0458

2017 A group decision making support system in logistics and supply chain management. Expert Systems with Applications, Elsevier (2017) 88, pp. 376-392, with Morteza Yazdani, Adama Coulibaly, Edmundas Kazimieras Zavadskas, DOI: 10.1016/j.eswa.2017.07.014

2016 A new trend for knowledge-based decision support systems design. International Journal of Information and Decision Sciences, InderScience (2016) 8(3), pp. 305-324, with Shaofeng Liu. 2013 Tools for collaborative decision-making. John Wiley (2013)

Séminaire DIC-ISC-CRIA – 24 mars 2022 avec Jean-Pierre NADAL et Laurent BONNASSE-GAHOT

Jean-Pierre NADAL et Laurent BONNASSE-GAHOT – 24 mars 2022

Titre : Perception catégorielle et géométrie des représentations internes d’un réseau de neurones artificiels

Résumé:

Usant d’outils mathématiques issus de la théorie de l’information, nous présentons un ensemble de résultats sur la modélisation des bases neuronales de la perception catégorielle, caractérisée par une compression intra-catégorielle et une séparation inter-catégorielle. Nous montrons notamment comment la perception catégorielle émerge naturellement de l’apprentissage de catégories. Ces résultats nous donnent des pistes pour l’analyse des représentations internes des réseaux de neurones artificiels. Un résultat important est l’analyse de l’interaction entre la géométrie et le bruit au cours de l’apprentissage.

Bio:

Ingénieur diplômé de Telecom ParisTech, avec une spécialisation en apprentissage machine, Laurent BONNASSE-GAHOT a ensuite complété son parcours par un master en sciences cognitives à l’École Normale Supérieure, puis par un doctorat à l’EHESS. Il travaille à présent au CAMS (CNRS-EHESS), en tant qu’ingénieur de recherche en analyse de données, avec comme intérêts de recherche les réseaux de neurones (naturels et artificiels), le traitement de la musique et du langage.

Jean-Pierre NADAL est Directeur de recherche au CNRS et Directeur d’études à l’Ecole des Hautes Etudes en Sciences Sociales (EHESS). Il partage sa recherche entre le Laboratoire de Physique de l’Ecole Normale Supérieure (LPENS, ENS – Université PSL – CNRS – SU – Université Paris Cité), et le Centre d’Analyse et de Mathématique Sociales (CAMS, CNRS – EHESS). Il est actuellement directeur du CAMS.

Ingénieur diplômé de l’Ecole Polytechnique, docteur en Physique Statistique, ses domaines de recherche sont, d’une part, les neurosciences computationnelles – et plus largement les sciences cognitives – et l’apprentissage machine, et, d’autre part, la modélisation de systèmes complexes en sciences économiques et sociales.

Séminaire DIC-ISC-CRIA – 17 mars 2022 avec Abbas GHADDAR

Abbas GHADDAR – 17 mars 2022

Titre : Context-aware Adversarial Training for Name Regularity Bias in Named Entity Recognition

Résumé:

In this work, we examine the ability of NER models to use contextual information when predicting the type of an ambiguous entity. We introduce NRB, a new testbed carefully designed to diagnose Name Regularity Bias of NER models. Our results indicate that all state-of-the-art models we tested show such a bias; BERT fine-tuned models significantly outperforming feature-based (LSTM-CRF) ones on NRB, despite having comparable (sometimes lower) performance on standard benchmarks.To mitigate this bias, we propose a novel model-agnostic training method that adds learnable adversarial noise to some entity mentions, thus enforcing models to focus more strongly on the contextual signal, leading to significant gains on NRB. Combining it with two other training strategies, data augmentation and parameter freezing, leads to further gains.

Bio:

Abbas GHADDAR is a Senior Researcher at Huawei. Abbas received his Master’s and Ph.D. degrees in 2016 and 2020, respectively, from the University of Montreal under the supervision of Professor Philippe Langlais. Since 2020, he works as a researcher at Huawei Noah Ark Lab located in Montreal. His main research interest is machine learning methods applied to natural language processing, model compression, robustness, and generalization. He has publications at top conferences and journals, including ACL, ACL, EMNLP, CoNLL, COLING, NeruIPS.

Séminaire DIC-ISC-CRIA – 3 mars 2022 avec Vimla L. PATEL

Vimla PATEL – 3 mars 2022

Titre : Biomedical Informatics and the Science of Cognition

Résumé:

My laboratory’s multidisciplinary research on medical cognition has shown the considerable importance of cognitive factors that determine how health professionals understand information, solve problems, use decision-support tools, and make decisions. These investigations into the process of diagnostic reasoning have made contributions to the design of today’s clinical AI systems. My talk will elaborate current cognitive-related research in biomedical informatics, with a focus on clinical aspects, building on earlier investigations to elucidate key lessons and challenges for the development of usable, useful, and safe decision-support systems that can augment human intelligence in the clinical world.

Bio:

Vimla L Patel is a Senior Research Scientist and Director of the Center for Cognitive Studies in Medicine and Public Health at the New York Academy of Medicine. A graduate of McGill University in Montreal, she was a Professor of Medicine and Psychology and Director of McGill Cognitive Science CenterHer early research related to cognitive mechanisms underlying expertise and medical decision-making. Her studies over the past two decades are on decision support technology and errors in complex clinical environments, addressing the role of cognition in biomedical informatics (human-technology interaction, cognitive design, distributed cognition, and team decision making) for a safer clinical workplace.

Séminaire DIC-ISC-CRIA – 24 février 2022 avec Essam MANSOUR

Essam MANSOUR – 24 février 2022

Titre : Towards Cognitive Data Science Platforms: Challenges and Opportunities

Résumé:

Similar to Open Data initiatives, data science as a community has launched initiatives for sharing not only data but entire pipelines, derivatives, artifacts, etc. (Open Data Science). However, the few efforts that exist focus on the technical part on how to facilitate sharing, conversion, etc. In this talk, we introduce a framework to help the scientific community to discover and learn from each other’s work automatically. One of the key concepts to enable our framework is to abstract from syntactical differences of existing platforms and instead focus on the semantics of datasets, artifacts, and data science pipelines. Once we understand the semantics, we can more easily identify similar or matching artifacts and combine them in a federated manner. We use knowledge graph technologies to retain a maximal degree of flexibility by capturing metadata and semantics in a flexible graph format. Our framework enables scientists to collaborate more effectively regardless of the data science platforms they use and encourages innovative applications to automate several aspects of data science based on the most recent data science experimentation. We support this automation by enabling deep learning models on our data science knowledge graph. The development of our framework poses numerous open research challenges that require innovative methodologies such as i) semantic pipeline abstraction, ii) exploring our data science knowledge graph using natural language questions, and iii) learning from decentralized knowledge graphs of different data science projects.

Bio:

Essam Mansour has been an assistant professor since 2019 in the Department of Computer-Science and Software Engineering (CSSE) at Concordia University in Montreal, and the head of the Cognitive Data Science (CoDS) lab. His research program focuses on developing Cognitive Data Science Platforms for federated and big datasets. His research interests are in the broad areas of knowledge graphs, distributed data systems, and graph neural networks. Essam spent more than 10 years doing world-class research, in the areas of databases, parallel/distributed systems, big data analytics, and querying geo-distributed graphs. He is developing and optimizing data science systems to work at scale on supercomputers and cloud resources. During these years, his research contributions have led to more than 30 conference and journal papers (mostly in top-tier venues, such as VLDBJ, PVLDB, SIGMOD, ICDE, EDBT, and CIKM). He has been invited as a reviewer for top journals, such as ACM Transactions on Database Systems (TODS), VLDB Journal, and IEEE Transactions on Knowledge and Data Engineering (TKDE). Essam is a meta-reviewer at SIGMOD 2023 and also has served as a program committee member in several top conferences, such as SIGMOD 2021, VLDB 2016 to 2021, SIGMOD 2016, and ICDE 2016.

Séminaire DIC-ISC-CRIA – 17 février 2022 avec Yoshua BENGIO

Yoshua BENGIO – 17 février 2022

Titre : Conscious processing, inductive biases and generalization in deep learning

Résumé:

Humans are very good at “out-of-distribution” generalization (compared to current AI systems). It would be useful to determine the inductive biases they exploit and translate them into machine-language architectures, training frameworks and experiments. I will discuss several of these hypothesized inductive biases. Many exploit notions in causality and connect abstractions in representation learning (perception and interpretation) with reinforcement learning (abstract actions). Systematic generalizations may arise from efficient factorization of knowledge into recomposable pieces. This is partly related to symbolic AI (aas seen in the errors and limitations of reasoning in humans, as well as in our ability to learn to do this at scale, with distributed representations and efficient search). Sparsity of the causal graph and locality of interventions — observable in the structure of sentences — may reduce the computational complexity of both inference (including planning) and learning. This may be why evolution incorporated this as « consciousness.” I will also suggest some open research questions to stimulate further research and collaborations.

Bio:

Yoshua Bengio est professeur titulaire à l’Université de Montréal, fondateur et directeur scientifique de Mila – Institut québécois d’IA, et codirige le programme Apprentissage automatique, apprentissage biologique de CIFAR en tant que Senior Fellow. Il occupe également la fonction de directeur scientifique d’IVADO

Suivez-nous