Guillaume Lajoie

Biographie

Guillaume Lajoie est professeur agrégé au Département de mathématiques et de statistiques (DMS) de l'Université de Montréal et membre académique principal de Mila – Institut québécois d’intelligence artificielle. Il est titulaire d'une chaire CIFAR (CCAI Canada) ainsi que d'une chaire de recherche du Canada (CRC) en calcul et interfaçage neuronaux.

Ses recherches sont positionnées à l'intersection de l'IA et des neurosciences où il développe des outils pour mieux comprendre les mécanismes d'intelligence communs aux systèmes biologiques et artificiels. Les contributions de son groupe de recherche vont des progrès des paradigmes d'apprentissage à plusieurs échelles pour les grands systèmes artificiels aux applications en neurotechnologie. Dr. Lajoie participe activement aux efforts de développement responsables de l'IA, cherchant à identifier les lignes directrices et les meilleures pratiques pour l'utilisation de l'IA dans la recherche et au-delà.

Étudiants actuels

Federico Arangath Joseph

Collaborateur·rice de recherche - ETH Zurich

Stefan Bauer

Visiteur de recherche indépendant

Superviseur⋅e principal⋅e :

Yoshua Bengio

Sangnie Bhardwaj

Doctorat - UdeM

Co-superviseur⋅e :

Hugo Larochelle

Colin Bredenberg

Postdoctorat - UdeM

Co-superviseur⋅e :

Leo Choiniere

Doctorat - UdeM

Olivier Codol

Postdoctorat - UdeM

Co-superviseur⋅e :

Matt Perich

Eric Elmoznino

Doctorat - UdeM

Superviseur⋅e principal⋅e :

Doctorat - UdeM

Superviseur⋅e principal⋅e :

Leo Gagnon

Doctorat - UdeM

Juan Guerra

Maîtrise recherche - Polytechnique

Superviseur⋅e principal⋅e :

Marco Bonizzato

Site web

Nanda Harishankar Krishna

Doctorat - UdeM

Collaborateur·rice de recherche - Western Washington University (faculty; assistant prof))

Superviseur⋅e principal⋅e :

Maîtrise recherche - UdeM

Co-superviseur⋅e :

Collaborateur·rice de recherche - UdeM

Ximeng Mao

Doctorat - UdeM

Co-superviseur⋅e :

Joelle Pineau

Abdel Mfougouon Njupoun

Doctorat - UdeM

Co-superviseur⋅e :

Doctorat - UdeM

Co-superviseur⋅e :

Amine Natik

Doctorat - UdeM

Co-superviseur⋅e :

Guy Wolf

François Paugam

Doctorat - UdeM

Superviseur⋅e principal⋅e :

Lune Bellec

Alexandre Payeur

Collaborateur·rice de recherche - UdeM

Mohammad Pezeshki

Collaborateur·rice de recherche

Superviseur⋅e principal⋅e :

Postdoctorat - McGill

Superviseur⋅e principal⋅e :

Param Raval

Collaborateur·rice alumni - UdeM

Avery Ryoo

Maîtrise recherche - UdeM

Superviseur⋅e principal⋅e :

Doctorat - UdeM

Co-superviseur⋅e :

Lune Bellec

Laura Suarez

Doctorat - McGill

Ryan Vogt

Postdoctorat - UdeM

Que nous apprennent les distributions des coefficients synaptiques au sujet de l’apprentissage dans le cerveau ?

Vivian White

Stagiaire de recherche - Western Washington University

Co-superviseur⋅e :

Doctorat - UdeM

Billets de blogue

13 juin 2024

par

Roman Pogodin

Jonathan Cornford

Arna Ghosh

Gauthier Gidel

Guillaume Lajoie

Blake Richards

Lire l'article

Publications

Learning Versatile Optimizers on a Compute Diet

Abhinav Moudgil

Boris Knyazev

Eugene Belilovsky

Learned optimization has emerged as a promising alternative to hand-crafted optimizers, with the potential to discover stronger learned upda… (voir plus)te rules that enable faster, hyperparameter-free training of neural networks. A critical element for practically useful learned optimizers, that can be used off-the-shelf after meta-training, is strong meta-generalization: the ability to apply the optimizers to new tasks. Recent state-of-the-art work in learned optimizers, VeLO (Metz et al., 2022), requires a large number of highly diverse meta-training tasks along with massive computational resources, 4000 TPU months, to achieve meta-generalization. This makes further improvements to such learned optimizers impractical. In this work, we identify several key elements in learned optimizer architectures and meta-training procedures that can lead to strong meta-generalization. We also propose evaluation metrics to reliably assess quantitative performance of an optimizer at scale on a set of evaluation tasks. Our proposed approach, Celo, makes a significant leap in improving the meta-generalization performance of learned optimizers and also outperforms tuned state-of-the-art optimizers on a diverse set of out-of-distribution tasks, despite being meta-trained for just 24 GPU hours.

2025-01-22

ArXiv (prépublication)

The oneirogen hypothesis: modeling the hallucinatory effects of classical psychedelics in terms of replay-dependent plasticity mechanisms

Colin Bredenberg

Fabrice Normandin

Classical psychedelics induce complex visual hallucinations in humans, generating percepts that are co-herent at a low level, but which have… (voir plus) surreal, dream-like qualities at a high level. While there are many hypotheses as to how classical psychedelics could induce these effects, there are no concrete mechanistic models that capture the variety of observed effects in humans, while remaining consistent with the known pharmacological effects of classical psychedelics on neural circuits. In this work, we propose the “oneirogen hypothesis”, which posits that the perceptual effects of classical psychedelics are a result of their pharmacological actions inducing neural activity states that truly are more similar to dream-like states. We simulate classical psychedelics’ effects via manipulating neural network models trained on perceptual tasks with the Wake-Sleep algorithm. This established machine learning algorithm leverages two activity phases, a perceptual phase (wake) where sensory inputs are encoded, and a generative phase (dream) where the network internally generates activity consistent with stimulus-evoked responses. We simulate the action of psychedelics by partially shifting the model to the ‘Sleep’ state, which entails a greater influence of top-down connections, in line with the impact of psychedelics on apical dendrites. The effects resulting from this manipulation capture a number of experimentally observed phenomena including the emergence of hallucinations, increases in stimulus-conditioned variability, and large increases in synaptic plasticity. We further provide a number of testable predictions which could be used to validate or invalidate our oneirogen hypothesis.

2025-01-13

bioRxiv (prépublication)

Brain-like learning with exponentiated gradients

Jonathan Cornford

Roman Pogodin

Arna Ghosh

Kaiwen Sheng

Brendan A. Bicknell

Olivier Codol

Beverley A. Clark

2024-10-26

bioRxiv (prépublication)

Multi-agent cooperation through learning-aware policy gradients

Alexander Meulemans

Seijin Kobayashi

Johannes von Oswald

Nino Scherrer

Eric Elmoznino

Blaise Agüera y Arcas

João Sacramento

Self-interested individuals often fail to cooperate, posing a fundamental challenge for multi-agent learning. How can we achieve cooperation… (voir plus) among self-interested, independent learning agents? Promising recent work has shown that in certain tasks cooperation can be established between learning-aware agents who model the learning dynamics of each other. Here, we present the first unbiased, higher-derivative-free policy gradient algorithm for learning-aware reinforcement learning, which takes into account that other agents are themselves learning through trial and error based on multiple noisy trials. We then leverage efficient sequence models to condition behavior on long observation histories that contain traces of the learning dynamics of other agents. Training long-context policies with our algorithm leads to cooperative behavior and high returns on standard social dilemmas, including a challenging environment where temporally-extended action coordination is required. Finally, we derive from the iterated prisoner's dilemma a novel explanation for how and when cooperation arises among self-interested learning-aware agents.

2024-10-24

ArXiv (prépublication)

A Complexity-Based Theory of Compositionality

Eric Elmoznino

Thomas Jiralerspong

Yoshua Bengio

2024-10-18

ArXiv (prépublication)

A Complexity-Based Theory of Compositionality

Eric Elmoznino

Thomas Jiralerspong

Yoshua Bengio

Compositionality is believed to be fundamental to intelligence. In humans, it underlies the structure of thought, language, and higher-level… (voir plus) reasoning. In AI, compositional representations can enable a powerful form of out-of-distribution generalization, in which a model systematically adapts to novel combinations of known concepts. However, while we have strong intuitions about what compositionality is, there currently exists no formal definition for it that is measurable and mathematical. Here, we propose such a definition, which we call representational compositionality, that accounts for and extends our intuitions about compositionality. The definition is conceptually simple, quantitative, grounded in algorithmic information theory, and applicable to any representation. Intuitively, representational compositionality states that a compositional representation satisfies three properties. First, it must be expressive. Second, it must be possible to re-describe the representation as a function of discrete symbolic sequences with re-combinable parts, analogous to sentences in natural language. Third, the function that relates these symbolic sequences to the representation, analogous to semantics in natural language, must be simple. Through experiments on both synthetic and real world data, we validate our definition of compositionality and show how it unifies disparate intuitions from across the literature in both AI and cognitive science. We also show that representational compositionality, while theoretically intractable, can be readily estimated using standard deep learning tools. Our definition has the potential to inspire the design of novel, theoretically-driven models that better capture the mechanisms of compositional thought.

2024-10-18

ArXiv (prépublication)

Learning Stochastic Rainbow Networks

Vivian White

Muawiz Sajjad Chaudhary

Guy Wolf

Kameron Decker Harris

Random feature models are a popular approach for studying network learning that can capture important behaviors while remaining simpler than… (voir plus) traditional training. Guth et al. [2024] introduced “rainbow” networks which model the distribution of trained weights as correlated random features conditioned on previous layer activity. Sampling new weights from distributions fit to learned networks led to similar performance in entirely untrained networks, and the observed weight covariance were found to be low rank. This provided evidence that random feature models could be extended to some networks away from initialization, but White et al. [2024] failed to replicate their results in the deeper ResNet18 architecture. Here we ask whether the rainbow formulation can succeed in deeper networks by directly training a stochastic ensemble of random features, which we call stochastic rainbow networks. At every gradient descent iteration, new weights are sampled for all intermediate layers and features aligned layer-wise. We find: (1) this approach scales to deeper models, which outperform shallow networks at large widths; (2) ensembling multiple samples from the stochastic model is better than retraining the classifier head; and (3) low-rank parameterization of the learnable weight covariances can approach the accuracy of full-rank networks. This offers more evidence for rainbow and other structured random feature networks as reduced models of deep learning.

2024-10-10

NeurIPS.cc/2024/Workshop/SciForDL (poster)

openreview.net

The oneirogen hypothesis: modeling the hallucinatory effects of classical psychedelics in terms of replay-dependent plasticity mechanisms

Colin Bredenberg

Fabrice Normandin

2024-09-30

bioRxiv (prépublication)

Accelerating Training with Neuron Interaction and Nowcasting Networks

Boris Knyazev

Abhinav Moudgil

Eugene Belilovsky

Simon Lacoste-Julien

Neural network training can be accelerated when a learnable update rule is used in lieu of classic adaptive optimizers (e.g. Adam). However,… (voir plus) learnable update rules can be costly and unstable to train and use. A simpler recently proposed approach to accelerate training is to use Adam for most of the optimization steps and periodically, only every few steps, nowcast (predict future) parameters. We improve this approach by Neuron interaction and Nowcasting (NiNo) networks. NiNo leverages neuron connectivity and graph neural networks to more accurately nowcast parameters by learning in a supervised way from a set of training trajectories over multiple tasks. We show that in some networks, such as Transformers, neuron connectivity is non-trivial. By accurately modeling neuron connectivity, we allow NiNo to accelerate Adam training by up to 50\% in vision and language tasks.

2024-09-06

ArXiv (prépublication)

Accelerating Training with Neuron Interaction and Nowcasting Networks

Boris Knyazev

Abhinav Moudgil

Eugene Belilovsky

Simon Lacoste-Julien

Neural network training can be accelerated when a learnable update rule is used in lieu of classic adaptive optimizers (e.g. Adam). However,… (voir plus) learnable update rules can be costly and unstable to train and use. Recently, Jang et al. (2023) proposed a simpler approach to accelerate training based on weight nowcaster networks (WNNs). In their approach, Adam is used for most of the optimization steps and periodically, only every few steps, a WNN nowcasts (predicts near future) parameters. We improve WNNs by proposing neuron interaction and nowcasting (NiNo) networks. In contrast to WNNs, NiNo leverages neuron connectivity and graph neural networks to more accurately nowcast parameters. We further show that in some networks, such as Transformers, modeling neuron connectivity accurately is challenging. We address this and other limitations, which allows NiNo to accelerate Adam training by up to 50% in vision and language tasks.

2024-09-06

ArXiv (prépublication)

When can transformers compositionally generalize in-context?

Seijin Kobayashi

Simon Schug

Yassir Akram

Florian Redhardt

Johannes von Oswald

Razvan Pascanu

João Sacramento

Many tasks can be composed from a few independent components. This gives rise to a combinatorial explosion of possible tasks, only some of w… (voir plus)hich might be encountered during training. Under what circumstances can transformers compositionally generalize from a subset of tasks to all possible combinations of tasks that share similar components? Here we study a modular multitask setting that allows us to precisely control compositional structure in the data generation process. We present evidence that transformers learning in-context struggle to generalize compositionally on this task despite being in principle expressive enough to do so. Compositional generalization becomes possible only when introducing a bottleneck that enforces an explicit separation between task inference and task execution.

2024-07-17

ArXiv (prépublication)

A benchmark of individual auto-regressive models in a massive fMRI dataset

Fraçois Paugam

Basile Pinsard

Lune Bellec

Abstract Dense functional magnetic resonance imaging datasets open new avenues to create auto-regressive models of brain activity. Individua… (voir plus)l idiosyncrasies are obscured by group models, but can be captured by purely individual models given sufficient amounts of training data. In this study, we compared several deep and shallow individual models on the temporal auto-regression of BOLD time-series recorded during a natural video-watching task. The best performing models were then analyzed in terms of their data requirements and scaling, subject specificity, and the space-time structure of their predicted dynamics. We found the Chebnets, a type of graph convolutional neural network, to be best suited for temporal BOLD auto-regression, closely followed by linear models. Chebnets demonstrated an increase in performance with increasing amounts of data, with no complete saturation at 9 h of training data. Good generalization to other kinds of video stimuli and to resting-state data marked the Chebnets’ ability to capture intrinsic brain dynamics rather than only stimulus-specific autocorrelation patterns. Significant subject specificity was found at short prediction time lags. The Chebnets were found to capture lower frequencies at longer prediction time lags, and the spatial correlations in predicted dynamics were found to match traditional functional connectivity networks. Overall, these results demonstrate that large individual functional magnetic resonance imaging (fMRI) datasets can be used to efficiently train purely individual auto-regressive models of brain activity, and that massive amounts of individual data are required to do so. The excellent performance of the Chebnets likely reflects their ability to combine spatial and temporal interactions on large time scales at a low complexity cost. The non-linearities of the models did not appear as a key advantage. In fact, surprisingly, linear versions of the Chebnets appeared to outperform the original non-linear ones. Individual temporal auto-regressive models have the potential to improve the predictability of the BOLD signal. This study is based on a massive, publicly-available dataset, which can serve for future benchmarks of individual auto-regressive modeling.

2024-07-01

Imaging Neuroscience (publié)