La prochaine rencontre, qui aura lieu le 10 novembre à Mila, permettra d'explorer comment pouvons-nous collectivement développer, encadrer et déployer des systèmes agentiques performants, fiables et sécuritaires en connectant chercheur·euse·s académiques, expert·e·s industriel·le·s et praticien·ne·s.
Les inscriptions sont ouvertes pour notre programme de 4 mois à temps plein, à Mila, conçu spécifiquement pour les fondateurs et fondatrices de la deep tech issus de parcours d'élite en STIM. Vous avez jusqu'au 21 novembre 2026 pour appliquer!
Le processus de demandes de supervision à Mila est maintenant ouvert! Vous pouvez appliquer dès maintenant à la maitrise de recherche ou Doctorat et joindre notre grande communauté de chercheur·euse·s à Mila pour l’automne 2027. Date limite : 1er décembre 2026
Nous utilisons des témoins pour analyser le trafic et l’utilisation de notre site web, afin de personnaliser votre expérience. Vous pouvez désactiver ces technologies à tout moment, mais cela peut restreindre certaines fonctionnalités du site. Consultez notre Politique de protection de la vie privée pour en savoir plus.
Paramètre des cookies
Vous pouvez activer et désactiver les types de cookies que vous souhaitez accepter. Cependant certains choix que vous ferez pourraient affecter les services proposés sur nos sites (ex : suggestions, annonces personnalisées, etc.).
Cookies essentiels
Ces cookies sont nécessaires au fonctionnement du site et ne peuvent être désactivés. (Toujours actif)
Cookies analyse
Acceptez-vous l'utilisation de cookies pour mesurer l'audience de nos sites ?
Lecteur Multimédia
Acceptez-vous l'utilisation de cookies pour afficher et vous permettre de regarder les contenus vidéo hébergés par nos partenaires (YouTube, etc.) ?
Publications
Automatic Lesion Morphometrics Reliably Predict Sensorimotor Recovery After Spinal Cord Injury: A Longitudinal, Multicenter Neuroimaging Study
OBJECTIVE: To evaluate automatic magnetic resonance imaging (MRI)-derived lesion morphometrics against manual measurements and their prognos… (voir plus)tic value for characterizing distinct acute to chronic neurological recovery trajectories after spinal cord injury (SCI). METHODS: We studied a retrospective, multicenter cohort of cervical and thoracolumbar SCI patients who underwent routine clinical MRI at 3-4 weeks post-SCI and neurological examinations over 6 months. Lesion severity was quantified using both established manual measurements and a novel automatic approach. We assessed the between-method agreement of MRI-derived lesion morphometrics, including lesion length, lesion width, and tissue bridge width, and their ability to stratify patients according to 6-month motor and sensory outcomes. Data-driven prediction models were used to identify clinically meaningful lesion measures and their respective thresholds, differentiating subgroups of specific recovery profiles. RESULTS: Eighty-three patients from eight sites were included. Automatically derived lesion morphometrics showed moderate to strong agreement with established manual measures (Spearman's ρ: 0.540-0.821, p 0.001), capturing inter-individual differences in lesion severity and demonstrating comparable prognostic value. Lesion width represented the strongest predictor of 6-month neurological outcomes, splitting patients into recovery-specific subgroups. Patients with a smaller lesion width (threshold: 5.88-6.41 mm) demonstrated higher 6-month sensorimotor scores and superior recovery trajectories, while tissue bridges (threshold: 0.17 mm) further refined motor prediction-based stratification of more severely affected patients. INTERPRETATION: Automatic MRI measures of lesion severity hold promise as scalable, time-efficient in vivo biomarkers of SCI prognosis, with potential for clinical adoption to guide treatment decisions and multicenter trial planning.
2026-11-04
Annals of Clinical and Translational Neurology (publié)
Leading theories of cortical function posit that sensory processing involves a comparison between top-down predictions and bottom-up sensory… (voir plus) data. Tests of these theories have been hampered by experimental challenges related to simultaneously recording top-down and bottom-up signals in response to sensory stimuli that either do or do not match internally-generated predictions. Here, we conduct dual-plane 2-photon calcium imaging near-simultaneously from distal apical dendrites and somata of pyramidal neurons in each of 4 different regions of mouse visual cortex while the mice are presented with visual sequences that can either match or mismatch previously habituated visual stimuli. Specifically, we examine the responses in the distal apical dendrites and cell bodies of pyramidal neurons while mice view either movie clips that reverse unexpectedly or Gabor patch sequences that have orientations that do not match habituated patterns. We find that in response to unexpected reversals in natural movies, both the apical dendrites and somata show increased activity that decreases over repeated exposures, consistent with a prediction error response. However, we do not find any clear evidence for increased activity in response to mismatches in Gabor sequences, beyond possible release from adaptation effects. We additionally use volumetric images to connect dendritic and somatic compartments and examine the coupling between connected dendrites and cell bodies in layer 5 pyramidal neurons. We do not find evidence for changes in coupling in response to mismatch stimuli, either for natural movies or Gabor sequences. Our results are consistent across four regions of visual cortex. We also observe that increased responses to reversals in natural movies are stronger in layer 2/3 than layer 5 pyramidal neurons. Altogether, these data suggest that visual cortex is sensitive to deviations from previously-experienced naturalistic inputs. Our data do not suggest that this mismatch sensitivity extends to artificial stimuli, nor do they support theories of prediction error causing changes in the coupling between apical dendrites and somata.
Unsupervised Environment Design (UED) approaches train reinforcement learning (RL) agents on diverse curricula that may evolve with their ca… (voir plus)pabilities. While prior work has largely focused on how these curricula are constructed, we study whether their students remain trainable under the resulting distributions. Across multiple UED mechanisms, environment complexities, and extended training horizons, we identify recurring learner-side pathologies, including parameter growth, degraded value--outcome ordering, and loss of representational diversity, that accompany stagnation and performance degradation. These findings connect limitations in UED to optimization failures previously identified in conventional deep RL, while showing how adaptive curricula couple the learner's optimization state to its future training distribution. Controlled interventions show that mitigating selected pathologies improves learning. Stabilizing the learner improves long-horizon performance and zero-shot generalization without changing the curriculum rule or increasing environment interaction. Our results establish learner trainability as a complementary design principle for UED. Effective curricula require students that remain capable of learning from the distributions they produce.
Neural networks are likely to be among the most consequential artefacts in human history. As these systems become more capable and autonomou… (voir plus)s, their potential to cause societal harm also grows, making AI safety one of the defining challenges of our time. Yet we continue to build increasingly powerful systems without fully understanding how they work, how they will behave in unfamiliar situations, or how to reliably control them. Mechanistic interpretability offers a path towards understanding these systems by studying their internal representations and computations, with the aim of evaluating, predicting, and controlling behaviour. We survey mechanistic interpretability for AI safety in order to answer the question: Does understanding how models work actually make them safer? We place mechanistic interpretability in the broader context of interpretability research and compare its mechanistic evidence with behavioural evaluation. We introduce the foundations of mechanistic interpretability, including features, polysemanticity, transformer computation, and circuits, and survey methods for feature extraction and analysis, circuit discovery and causal verification, explaining intermediate representations in natural language, learning dynamics, and training-data attribution. Beyond surveying existing methods, we propose a unified framework for evaluating mechanistic interpretability claims and characterise six research orientations by their goals, assumptions, and trade-offs. Finally, we examine how mechanistic understanding can support AI safety, including knowledge editing, steering, and monitoring, and assess its limitations, failure modes, open problems, and relationship to behavioural safety measures. We also release a repository of code notebooks with worked examples that reproduce key findings from surveyed methods, allowing readers to build intuition through hands-on experimentation.
We introduce dithered belief-propagation decoding for circuit-level quantum memory to mitigate the oscillations and trapping behavior that l… (voir plus)imit standard belief-propagation (BP) decoders for qLDPC codes. Our approach combines a quantum-adapted sign-flip mechanism with chained randomized prior perturbation. Evaluated under circuit-level noise on the distance-11 rotated surface code and the distance-12 bivariate-bicycle code, the proposed dithered decoder significantly outperforms BP and BP+OSD+CS-10. The proposed decoder achieves accuracy comparable to or better than state-of-the-art Relay-BP, especially at higher physical error rates, while trading a moderate increase in additions for about 98% fewer multiplications on average and eliminating node-level RNGs. These results suggest that dithering is a promising approach toward real-time decoding of quantum memories.
As agentic AI becomes more widespread, agents with distinct and possibly conflicting goals will interact in complex ways. These multi-agent … (voir plus)interactions pose a fundamental challenge, particularly in social dilemmas, where agents' individual incentives can undermine collective welfare. While reinforcement learning (RL) has been effective for aligning large language models (LLMs) in the single-agent regime, prior small-network results suggest that standard RL in multi-agent settings often converges to defecting, self-interested policies. We show the same effect in LLMs: despite cooperative priors, RL-trained LLM agents develop opportunistic behaviour that can exploit even advanced models. To address this tendency of RL to converge to poor equilibria, we adapt a recent opponent-learning awareness algorithm, Advantage Alignment, to fine-tune LLMs toward multi-agent cooperation and robustness against exploitation. We then introduce a group-relative baseline that simplifies advantage computation in iterated games, enabling multi-agent training at LLM scale. We also contribute a novel social dilemma environment, Trust-and-Split, which requires natural language communication to achieve high collective welfare. Across a wide range of social dilemmas, policies learned with Advantage Alignment achieve higher collective payoffs while remaining robust against exploitation by greedy agents.
2026-11-03
Transactions on Machine Learning Research (accepté)
Abstract The rapid rise of artificial intelligence (AI), especially generative AI, is transforming industries and redefining corporate gover… (voir plus)nance. AI now influences key business functions such as finance, compliance and risk management, presenting boards of directors with new oversight challenges. While AI offers strategic opportunities, it also introduces complex risks – algorithmic bias, privacy breaches, cybersecurity threats and ethical concerns. Despite its growing relevance, AI governance remains underdeveloped in Canada. From an empirical legal standpoint, the role of boards in managing AI-related risks is still emerging and largely unexplored. This gap is increasingly critical amid rising expectations from regulators, investors and the spread of soft law instruments like voluntary codes and proxy advisor guidelines. This article examines how Canadian TSX 60 companies are responding to AI governance challenges. It situates the issue within Canada’s corporate governance framework and presents an empirical study revealing a disconnect between recognizing AI as a strategic risk and formally integrating it into governance structures. Drawing on academic research and international best practices, the article assesses the responsiveness of Canadian governance standards and offers recommendations for regulators seeking to modernize oversight in an era of technological disruption and ethical complexity.
Self-attention has become the standard for modern sequence modeling, enabling Transformers to model long-range dependencies across a wide ra… (voir plus)nge of modalities. Despite these advances, its quadratic cost with respect to sequence length remains a significant computational bottleneck. Recent work addresses this limitation by introducing linear variants of self-attention, namely with kernel approximations of the softmax operator. However, their first-order aggregation restricts expressivity and struggles to capture higher-order structure. We propose Linear Stochastic Attention (LiSA), a linear attention plug-in module that enables higher-order token interactions via a stochastic diffusion process based on truncated random walks. LiSA incorporates adaptive locality, favoring semantically similar neighbors to preserve fine-grained structures, while maintaining stable and efficient training. We conduct extensive experiments on classification, semantic segmentation, and object detection tasks and show improved results over state-of-the-art methods.
Abstract Population genetic data are increasingly visualized using low-dimensional embeddings as ancestry coordinates, yet how these methods… (voir plus) represent population structure can depend strongly on sampling and construction choices. Principal component analysis (PCA), when visualized using only leading components, can obscure fine-scale structure, while nonlinear neighbor-based methods such as uniform manifold approximation and projection (UMAP) often fragment geographic and admixture gradients into discrete clusters. Here, we introduce a framework for constructing ancestry coordinates using potential of heat-diffusion for affinity-based trajectory embedding (PHATE), a diffusion-based method that preserves relationships across scales. Across synthetic benchmarks and large genomic datasets, we show that appropriate scale tuning and population balancing allow PHATE to recover geographic organization and admixture gradients. We show that incorrectly tuned neighborhood scale and population imbalance can distort the resulting representations of population structure. These results therefore provide a methodological framework and practical guidance for constructing ancestry visualizations that better reflect continuous human genetic diversity.
Moving robot swarms from simulation into the physical world can be difficult because small discrepancies in hardware, timing and communicati… (voir plus)on can cascade across interacting robots and alter group behaviour. This challenge is known as the reality gap. When swarms operate in human environments, a second challenge emerges: a perceptual gap between what the swarm physically does and what human onlookers actually perceive. This study examines the same decentralised coordination code deployed across three robotic swarms-tabletop, aerial and human-scale mobile manipulators-during live artistic performances. We find that while the underlying algorithms executed reliably on all three platforms, physical embodiment drastically altered visible motion-distorting execution timescales by more than a factor of three and reshaping trajectory geometry. More importantly, audience surveys revealed that human interpretation does not mirror technical success: how observers attribute meaning (perceived purpose) and how they react emotionally (comfort versus threat) depend heavily on physical form. Meaning attribution and emotional responses did not vary together. Our results demonstrate that successfully transferring swarm software to physical hardware is insufficient for real-world deployment. Addressing the reality gap does not guarantee that the resulting collective motion will be readily interpreted by human observers. Swarms destined for human environments must be engineered not only for decentralised stability but also for the expressive physical dynamics through which humans interpret collective motion.
Despite their widespread success across machine learning, the learning dynamics and generalization properties of neural networks remain poor… (voir plus)ly understood. The neural tangent kernel (NTK) provides an analytical characterization of the learning dynamics of infinitely wide neural networks and serves as a useful proxy for finite-width networks in supervised learning. In this work, we extend the NTK framework to reinforcement learning (RL) by introducing a temporal-difference (TD) flow formulation that accommodates time-varying mini-batches and TD targets, yielding a closed-form solution for value function evolution in an overparametrized setting. We also devise an optimization-based state-space exploration method that leverages the closed-form solution, coupled with constraints on the environment, to model the set of feasible states. On a deterministic grid environment, we validate that the proposed RL-based NTK formulation is indeed learning efficiently and adapting to the distribution shift in the training data distribution, and we demonstrate that the resulting exploration strategy improves sample efficiency. To our knowledge, this is the first work to extend NTK theory to characterize the learning dynamics of RL training with time-varying mini-batches and TD updates.
Abstract Airborne tree monitoring using remotely sensed data, combined with the growth of computer vision models for object detection and se… (voir plus)gmentation, can facilitate broad scale assessment of tree health, ecology and ecosystem function. However, datasets for airborne tree detection have been largely limited to a single geography, resolution, or annotation approach, making it difficult to develop general models and benchmark their utility for broad applications. We compile existing and newly published annotation datasets to create a comprehensive dataset for training and evaluating airborne tree detection in RGB imagery. The MillionTrees dataset has over a million individual trees from a broad range of forest types, tree species, and geographic locations. In addition, it has six million weakly supervised annotations from a range of sources including airborne LiDAR and photogrammetry, arborist surveys, and simulated landscapes. An evaluation standard is provided to facilitate rapid model development. We benchmark several existing tree detection algorithms and demonstrate that while current open-source models do not generalize across the diverse datasets in the benchmark, training on MillionTrees improves both within-domain and out-of-domain performance. Held-out validation datasets using ground truth derived from terrestrial LiDAR paired with overhead RGB show that overfitting to test data is common and hyperparameter tuning is an underappreciated challenge towards a global zero shot tree detection method. We discuss the limitations of the currently available data for airborne RGB tree detection, how to create more robust evaluation data and the need for better understanding of the uncertainty and utility of tree detection from airborne remote sensing.