Guillaume Lajoie

matteo.alleman@mila.quebec

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

Matteo Alleman

Postdoctorat - UdeM

Rohan Banerjee

Collaborateur·rice alumni - Polytechnique

rohan.banerjee@mila.quebec

Visiteur de recherche indépendant - N/A

philippe.beaudoin@mila.quebec

Doctorat - UdeM

Co-superviseur⋅e :

Hugo Larochelle

sangnie.bhardwaj@mila.quebec

Colin Bredenberg

Visiteur de recherche indépendant - UdeM

Co-superviseur⋅e :

colin.bredenberg@mila.quebec

Leo Choiniere

Doctorat - UdeM

leo.choiniere@mila.quebec

Claudia Clopath

Visiteur de recherche indépendant - University College London

claudia.clopath@mila.quebec

Eric Elmoznino

Doctorat - UdeM

Superviseur⋅e principal⋅e :

Yoshua Bengio

eric.elmoznino@mila.quebec

Leo Gagnon

Doctorat - UdeM

leo.gagnon@mila.quebec

Tom George

Postdoctorat - McGill

Superviseur⋅e principal⋅e :

tomgeorge1@btinternet.com

Juan Guerra

Doctorat - Polytechnique

juan.guerra@mila.quebec

thomas.jiralerspong@mila.quebec

Nanda Harishankar Krishna

Doctorat - UdeM

Anna Jahn

Visiteur de recherche indépendant - McGill

anna.jahn1@mila.quebec

Chen Jiang

Doctorat - McGill

Superviseur⋅e principal⋅e :

Paul Masset

chen.jiang@mila.quebec

Thomas Jiralerspong

Doctorat - UdeM

Co-superviseur⋅e :

Yoshua Bengio

Maîtrise recherche - UdeM

Co-superviseur⋅e :

Dhanya Sridhar

tejaskasetty@gmail.com

Doctorat - McGill

Superviseur⋅e principal⋅e :

pingsheng.li@mila.quebec

mathys.loiselle@mila.quebec

Mathys Loiselle

Stagiaire de recherche - Concordia

Co-superviseur⋅e :

Matt Perich

Ximeng Mao

Doctorat - UdeM

Co-superviseur⋅e :

Joelle Pineau

ximeng.mao@mila.quebec

Abdel Mfougouon Njupoun

Doctorat - UdeM

Co-superviseur⋅e :

alexandre.payeur@mila.quebec

abdelnjupoun@gmail.com

Doctorat - UdeM

Co-superviseur⋅e :

Visiteur de recherche indépendant - UdeM

Julia Price

Maîtrise recherche - UdeM

julia.price@mila.quebec

Mauricio Rivera

Maîtrise recherche - UdeM

Superviseur⋅e principal⋅e :

Marco Bonizzato

mauriv2010@gmail.com

hee-woon.ryoo@mila.quebec

Avery Ryoo

Doctorat - UdeM

Superviseur⋅e principal⋅e :

Matt Perich

Doctorat - UdeM

Co-superviseur⋅e :

Lune Bellec

sainathp@mila.quebec

ezekiel.williams@mila.quebec

Ryan Vogt

Postdoctorat - UdeM

ryan.vogt@mila.quebec

Ezekiel Williams

Doctorat - UdeM

Visiteur de recherche indépendant - University of Western Ontario

jinkang.xiang@mila.quebec

Billets de blogue

Représentation graphique d'un nerf vague

21 mai 2025

Apprentissage automatique pour la segmentation des différentes activations des fibres nerveuses à partir des signaux neuronaux du cerveau vers le corps

par

Param Raval

Olivier Tessier-Larivière

Pascal Fortier-Poisson

Blake Richards

Guillaume Lajoie

Lire l'article

13 juin 2024

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

par

Roman Pogodin

Jonathan Cornford

Arna Ghosh

Gauthier Gidel

Guillaume Lajoie

Blake Richards

Lire l'article

Publications

In-context learning and Occam's razor

A central goal of machine learning is generalization. While the No Free Lunch Theorem states that we cannot obtain theoretical guarantees fo… (voir plus)r generalization without further assumptions, in practice we observe that simple models which explain the training data generalize best: a principle called Occam's razor. Despite the need for simple models, most current approaches in machine learning only minimize the training error, and at best indirectly promote simplicity through regularization or architecture design. Here, we draw a connection between Occam's razor and in-context learning: an emergent ability of certain sequence models like Transformers to learn at inference time from past observations in a sequence. In particular, we show that the next-token prediction loss used to train in-context learners is directly equivalent to a data compression technique called prequential coding, and that minimizing this loss amounts to jointly minimizing both the training error and the complexity of the model that was implicitly learned from context. Our theory and the empirical experiments we use to support it not only provide a normative account of in-context learning, but also elucidate the shortcomings of current in-context learning methods, suggesting ways in which they can be improved. We make our code available at https://github.com/3rdCore/PrequentialCode.

2025-10-05

Proceedings of the 42nd International Conference on Machine Learning (publié)

proceedings.mlr.press

Towards a Formal Theory of Representational Compositionality

2025-10-05

Proceedings of the 42nd International Conference on Machine Learning (publié)

proceedings.mlr.press

Recursive Self-Aggregation Unlocks Deep Thinking in Large Language Models

Johan Samir Obando Ceron

Yoshua Bengio

Brian R. Bartoldson

Bhavya Kailkhura

Glen Berseth

Nikolay Malkin

Moksh Jain

Test-time scaling methods improve the capabilities of large language models (LLMs) by increasing the amount of compute used during inference… (voir plus) to make a prediction. Inference-time compute can be scaled in parallel by choosing among multiple independent solutions or sequentially through self-refinement. We propose Recursive Self-Aggregation (RSA), a test-time scaling method inspired by evolutionary methods that combines the benefits of both parallel and sequential scaling. Each step of RSA refines a population of candidate reasoning chains through aggregation of subsets to yield a population of improved solutions, which are then used as the candidate pool for the next iteration. RSA exploits the rich information embedded in the reasoning chains -- not just the final answers -- and enables bootstrapping from partially correct intermediate steps within different chains of thought. Empirically, RSA delivers substantial performance gains with increasing compute budgets across diverse tasks, model families and sizes. Notably, RSA enables Qwen3-4B-Instruct-2507 to achieve competitive performance with larger reasoning models, including DeepSeek-R1 and o3-mini (high), while outperforming purely parallel and sequential scaling strategies across AIME-25, HMMT-25, Reasoning Gym, LiveCodeBench-v6, and SuperGPQA. We further demonstrate that training the model to combine solutions via a novel aggregation-aware reinforcement learning approach yields significant performance gains. Code available at https://github.com/HyperPotatoNeo/RSA.

2025-09-29

ArXiv (prépublication)

Inferring dynamical features from neural data through joint learning of latents factors and weights

Anirudh Gururaj Jamkhandi

Ali Korojy

Olivier Codol

Matthew G Perich

Behavior arises from coordinated synaptic changes in recurrent neural populations. Inferring the underlying dynamics from limited, noisy, an… (voir plus)d high-dimensional neural recordings is a major challenge, as experimental data often provide only partial access to brain states. While data-driven recurrent neural networks (dRNNs) have been effective for modeling such dynamics, they are typically limited to single-task domains and struggle to generalize across behavioral conditions. Here, we propose a hierachical model that captures neural dynamics across multiple behavioral contexts by learning a shared embedding space over RNN weights. We demonstrate that our model captures diverse neural dynamics with a single, unified model using both simulated datasets of many tasks and neural recordings during monkey reaching. Using the learned task embeddings, we demonstrate accurate classification of dynamical regimes and generalization to unseen samples. Crucially, spectral analysis on the learnt weights provide valuable insights into network computations, highlighting the potential of joint embedding–weight learning for scalable inference of brain dynamics.

2025-09-22

NeurIPS.cc/2025/Workshop/NeurReps (poster)

Towards a generalizable, unified framework for decoding from multimodal neural activity

Nanda H Krishna

Mathys Loiselle

Avery Hee-Woon Ryoo

Matthew G Perich

Recent advances in neural decoding have led to the development of large-scale deep learning-based neural decoders that can generalize across… (voir plus) sessions and subjects. However, existing approaches predominantly focus on single modalities of neural activity, limiting their applicability to specific modalities and tasks. In this work, we present a multimodal extension of the POYO framework that jointly processes neuronal spikes and local field potentials (LFPs) for behavioural decoding. Our approach employs flexible tokenization schemes for both spikes and LFPs, enabling efficient processing of heterogeneous neural populations without preprocessing requirements like binning. Through experiments on data from nonhuman primates performing motor tasks, we demonstrate that multimodal pretraining yields superior decoding performance compared to unimodal baselines. We also show evidence of cross-modal transfer: models pretrained on both modalities outperform LFP-only models when fine-tuned solely on LFPs, suggesting a path toward more cost-effective brain-computer interfaces that can use performant LFP-based decoders. Our models also exhibit robustness to missing modalities during inference when trained with modality masking, and scale effectively with both model size and pretraining data. Overall, this work represents an important first step towards unified, general-purpose neural decoders capable of leveraging diverse neural signals for a variety of brain-computer interface applications.

2025-09-21

NeurIPS.cc/2025/Workshop/BrainBodyFM (publié)

Celo: Training Versatile Learned Optimizers on a Compute Diet

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.

2025-07-07

TMLR (accepté)

Neuromorphic hierarchical modular reservoirs

Filip Milisav

Andrea I Luppi

Laura E Suarez

Bratislav Misic

Modularity is a fundamental principle of brain organization, reflected in the presence of segregated sub-networks that enable specialized in… (voir plus)formation processing. These small, densely connected modules are often nested within larger, higher-order modules, giving rise to a hierarchical modular architecture. This structure is posited to balance information segregation in specialized neuronal communities and global integration via intermodular communication. Yet, how hierarchical modularity shapes network function remains unclear. Here we introduce a simple blockmodeling framework for generating and comparing multi-level hierarchical modular networks and implement them as recurrent neural network reservoirs to evaluate their computational capacity. We show that hierarchical modular networks enhance memory capacity, support multitasking, and give rise to a broader range of temporal dynamics compared to strictly modular and random networks. These functional advantages can be traced to topological features enriched in hierarchical modular networks, which include reciprocal and cyclic network motifs. To test whether the computational advantages of hierarchical modularity subsist in empirical human brain structural connectivity patterns, we develop a novel hierarchical modularity-preserving network null model, allowing us to isolate the positive effect of empirical hierarchical modularity patterns on memory capacity. To evaluate the biomimetic validity of connectome-informed reservoir dynamics, we compare reservoir timescales to empirical brain timescales derived from MEG data and find that hierarchical modularity contributes to shaping brain-like neural timescales. Altogether, across multiple benchmarks, these results show that hierarchical modularity endows networks with computationally advantageous properties, providing insight into the relationship between neural network structure and function with potential applications for the design of neuromorphic computing architectures.

2025-06-20

bioRxiv (prépublication)

Next-Token Prediction Should be Ambiguity-Sensitive: A Meta-Learning Perspective

The rapid adaptation ability of auto-regressive foundation models is often attributed to the diversity of their pre-training data. This is b… (voir plus)ecause, from a Bayesian standpoint, minimizing prediction error in such settings requires integrating over all plausible latent hypotheses consistent with observations. While this behavior is desirable in principle, it often proves too ambitious in practice: under high ambiguity, the number of plausible latent alternatives makes Bayes-optimal prediction computationally intractable. Cognitive science has long recognized this limitation, suggesting that under such conditions, heuristics or information-seeking strategies are preferable to exhaustive inference. Translating this insight to next-token prediction, we hypothesize that low- and high-ambiguity predictions pose different computational demands, making ambiguity-agnostic next-token prediction a detrimental inductive bias. To test this, we introduce MetaHMM, a synthetic sequence meta-learning benchmark with rich compositional structure and a tractable Bayesian oracle. We show that Transformers indeed struggle with high-ambiguity predictions across model sizes. Motivated by cognitive theories, we propose a method to convert pre-trained models into Monte Carlo predictors that decouple task inference from token prediction. Preliminary results show substantial gains in ambiguous contexts through improved capacity allocation and test-time scalable inference, though challenges remain.

2025-06-18

ArXiv (prépublication)

Next-Token Prediction Should be Ambiguity-Sensitive : A Meta-Learing Perspective

2025-06-10

ICML.cc/2025/Workshop/ES-FoMo-III (publié)

MesaNet: Sequence Modeling by Locally Optimal Test-Time Training

Johannes Von Oswald

Nino Scherrer

Seijin Kobayashi

Luca Versari

Songlin Yang

Maximilian Schlegel

Kaitlin Maile

Yanick Schimpf

Oliver Sieberling

Alexander Meulemans

Rif A. Saurous

Charlotte Frenkel

Razvan Pascanu

Blaise Agüera y Arcas

João Sacramento

Sequence modeling is currently dominated by causal transformer architectures that use softmax self-attention. Although widely adopted, trans… (voir plus)formers require scaling memory and compute linearly during inference. A recent stream of work linearized the softmax operation, resulting in powerful recurrent neural network (RNN) models with constant memory and compute costs such as DeltaNet, Mamba or xLSTM. These models can be unified by noting that their recurrent layer dynamics can all be derived from an in-context regression objective, approximately optimized through an online learning rule. Here, we join this line of work and introduce a numerically stable, chunkwise parallelizable version of the recently proposed Mesa layer (von Oswald et al., 2024), and study it in language modeling at the billion-parameter scale. This layer again stems from an in-context loss, but which is now minimized to optimality at every time point using a fast conjugate gradient solver. Through an extensive suite of experiments, we show that optimal test-time training enables reaching lower language modeling perplexity and higher downstream benchmark performance than previous RNNs, especially on tasks requiring long context understanding. This performance gain comes at the cost of additional flops spent during inference time. Our results are therefore intriguingly related to recent trends of increasing test-time compute to improve performance -- here by spending compute to solve sequential optimization problems within the neural network itself.

2025-06-04

ArXiv (prépublication)

Bidirectional Information Flow (BIF) -- A Sample Efficient Hierarchical Gaussian Process for Bayesian Optimization

Juan D. Guerra

Thomas Garbay

Marco Bonizzato

Hierarchical Gaussian Process (H-GP) models divide problems into different subtasks, allowing for different models to address each part, mak… (voir plus)ing them well-suited for problems with inherent hierarchical structure. However, typical H-GP models do not fully take advantage of this structure, only sending information up or down the hierarchy. This one-way coupling limits sample efficiency and slows convergence. We propose Bidirectional Information Flow (BIF), an efficient H-GP framework that establishes bidirectional information exchange between parent and child models in H-GPs for online training. BIF retains the modular structure of hierarchical models - the parent combines subtask knowledge from children GPs - while introducing top-down feedback to continually refine children models during online learning. This mutual exchange improves sample efficiency, enables robust training, and allows modular reuse of learned subtask models. BIF outperforms conventional H-GP Bayesian Optimization methods, achieving up to 4x and 3x higher

2025-05-15

arXiv (Cornell University) (prépublication)

Does learning the right latent variables necessarily improve in-context learning?

Large autoregressive models like Transformers can solve tasks through in-context learning (ICL) without learning new weights, suggesting ave… (voir plus)nues for efficiently solving new tasks. For many tasks, e.g., linear regression, the data factorizes: examples are independent given a task latent that generates the data, e.g., linear coefficients. While an optimal predictor leverages this factorization by inferring task latents, it is unclear if Transformers implicitly do so or if they instead exploit heuristics and statistical shortcuts enabled by attention layers. Both scenarios have inspired active ongoing work. In this paper, we systematically investigate the effect of explicitly inferring task latents. We minimally modify the Transformer architecture with a bottleneck designed to prevent shortcuts in favor of more structured solutions, and then compare performance against standard Transformers across various ICL tasks. Contrary to intuition and some recent works, we find little discernible difference between the two; biasing towards task-relevant latent variables does not lead to better out-of-distribution performance, in general. Curiously, we find that while the bottleneck effectively learns to extract latent task variables from context, downstream processing struggles to utilize them for robust prediction. Our study highlights the intrinsic limitations of Transformers in achieving structured ICL solutions that generalize, and shows that while inferring the right latents aids interpretability, it is not sufficient to alleviate this problem.

2025-04-30

International Conference on Machine Learning (poster)