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
Scaling Density Functional Theory with Gaussian Splatting
Density functional theory (DFT) strikes a practical balance between accuracy and computational cost in many problems of computational chemis… (voir plus)try and materials science. However, many DFT calculations are limited by fixed atom-centered basis sets, which dictate how accuracy and cost scale with system size. We propose Gaussian Splatting for Density Functional Theory (GS-DFT), which represents molecular orbitals as a cloud of Gaussians whose positions, shapes, and mixing coefficients are optimized jointly by gradient descent to minimize the energy without training data. Conceptually, GS-DFT is 3D Gaussian splatting with the renderer replaced by quantum mechanics. We introduce two key solver components: adaptive density fitting with screening for efficient evaluation of two-electron integrals, and a regularized differentiable orthogonalization of the molecular orbitals. Empirically, the optimized basis reaches the accuracy of the largest conventional basis sets with a fraction of the parameters, converging systematically in energy, density, and nuclear forces. At equal parameter count, it captures the stretched-bond and anion physics that fixed bases only recover with specialized basis augmentation. The resulting solver exhibits quadratic peak memory scaling in the cloud size, allowing us to simulate systems of up to 2,742 atoms (10,406 electrons) without any modifications at triple-zeta scale using a single four-GPU node.
Mathematical expressions and the numerical behavior they produce are two views of the same underlying function, and connecting them is centr… (voir plus)al to scientific discovery. Symbolic Regression (SR) relies on this connection directly: it searches for an expression that reproduces a given behavior. Recent multi-modal models learn this connection by embedding symbolic expressions and their numerical behavior in a shared representation space. We show that this embedding space is only globally aligned: complete expressions correspond to complete behaviors, but the contribution of individual parts of an expression is not represented. This granularity gap leaves the model unable to tell how a local edit to an expression changes its behavior, the central operation in SR. We introduce a compositional alignment method that closes this gap: a structural positional encoding exposes the substructure of an expression to the encoder, and a multi-granularity contrastive objective grounds each subexpression in the behavior it produces before propagating this grounding to the full expression. The resulting representations close much of the modality gap between symbolic and numerical embeddings, reliably distinguish the effects of local edits that the original alignment cannot, and transfer to external SR corpora.
Language models are often evaluated on curated benchmarks that underrepresent the complexity of enterprise deployments. We introduce HARDEN,… (voir plus) a constrained evolutionary search method to adapt the input of existing evaluation cases into more challenging variants while keeping their expected outputs fixed. HARDEN searches along generated domain-specific complexity axes while enforcing feasibility constraints such as preserving task semantics, realism, and execution validity. Across FinQA, PubMedQA, and ContractNLI and three Qwen3.5 model scales (35B-A3B, 122B-A10B, and 397B-A17B), HARDEN reduces task-model accuracy by 22.7% on average and by up to 49.9% relative to single-pass baselines using the same feasibility checks. These results show that evolutionary search can produce substantially harder valid evaluation cases.
Language models are often evaluated on curated benchmarks that underrepresent the complexity of enterprise deployments. We introduce HARDEN,… (voir plus) a constrained evolutionary search method to adapt the input of existing evaluation cases into more challenging variants while keeping their expected outputs fixed. HARDEN searches along generated domain-specific complexity axes while enforcing feasibility constraints such as preserving task semantics, realism, and execution validity. Across FinQA, PubMedQA, and ContractNLI and three Qwen3.5 model scales (35B-A3B, 122B-A10B, and 397B-A17B), HARDEN reduces task-model accuracy by 22.7% on average and by up to 49.9% relative to single-pass baselines using the same feasibility checks. These results show that evolutionary search can produce substantially harder valid evaluation cases.
Introduction The motor cortex is organized to coordinate voluntary movements through distinct subregions that control different parts of the… (voir plus) body. Intracortical microstimulation (ICMS) with long-duration trains (LD-ICMS) can uncover complex, behaviorally relevant movement patterns that differ from the simple, joint-specific responses evoked by conventional short-duration trains (SD-ICMS). While previous studies have primarily examined forelimb representations, the organization of hindlimb movements remains less understood. Methods Here, we characterized hindlimb movements evoked by LD-ICMS, analyzing their kinematic features, cortical location and functional relevance in rats. Results Two distinct complex movements, Advance and Elevation, were identified, each with unique kinematic signatures. Principal component analysis revealed that overall movement amplitude and joint flexion patterns underlying movement trajectory systematically distinguished the two movement types. Advance and Elevation movements were spatially segregated within anteromedial and posterolateral cortical subregions, respectively. Phase-coherent stimulation of these subregions during locomotion selectively biased hindlimb trajectories forward or backward in an amplitude-dependent manner. Discussion These results demonstrate that the rat hindlimb motor cortex contains distinct subregions specialized for directionally specific and functionally relevant movement control.
With the gaining prominence of aerial mobility applications, their success depends on the seamless integration of terrestrial and non-terres… (voir plus)trial network connectivity. However, providing reliable connectivity from terrestrial telecommunication networks remains challenging due to multi-cell interference from base stations (BSs) under line-of-sight (LoS) conditions to unmanned aerial vehicles (UAVs), coverage holes caused by antenna sidelobe degradation, localized multipath fading effects, and the high-speed dynamics of aerial users. To model such complexities, often exacerbated by sparse real-world data, this work proposes a dual-stage radio environment map (REM) framework. Our approach physically decouples the channel modeling, where a spatial Transformer first anchors the deterministic, large-scale path loss geometry, while a gated recurrent unit (GRU) subsequently extrapolates the stochastic, localized fast- fading deviations. By reformulating 3D spatial interpolation as a 1D radial sequence prediction task, the framework inherently aligns with the physics of propagation. We evaluate the proposed framework against state-of-the-art baselines, including 3D Kriging, UNet, Mamba, and Inception, using empirical 5G datasets. The results demonstrate improved intra-site generalization across diverse altitudes, user dynamics, and reference signal received power (RSRP) datasets, achieving signal-strength predictions with errors near 3 dB and REM spatial-similarity indices exceeding 0.75. Finally, we examine the influence of REMs and channel rank conditions on UAV channel quality, underscoring the necessity of reliable channel modeling for robust aerial connectivity.
Navigation in off-road conditions is challenging due to the lack of structure. There is no fixed vocabulary for what is traversable. The tra… (voir plus)versability depends on both the environment and the embodiment's dynamics. Neither of these two variables can be hand-labeled at scale. Thus, traversability has to be learned by the embodiment's own experience. Modern platforms tend to use multiple sensors to estimate traversability and navigate: RGBD cameras, lidar, radar, IMU, with computationally intensive platforms to run inference on neural networks. Against this trend, we propose FLINT, a lightweight traversability estimator: a 21.6M-parameter backbone, 38\times smaller than a comparable foundation-model backbone, that scores higher on held-out terrain probes and runs at 14.7 FPS on CPU alone using a RGB camera has the only sensor. Despite that gap in scale, FLINT produces a cheaper, more accurate costmap than a deployed foundation-model system (WildOS) on 23 of 24 replayed field logs. We compare different self-supervised learning signals and deploy the resulting models on a real platform in closed-loop field trials: the best self-supervised head reaches 99% autonomy over the route, outperforming a human-label-trained baseline deployed live on the same course. Our results show that heavy sensing and computing are not necessary for traversability estimation.
We describe deploying imaging AI at six hospitals through PACS-AI, an open self-hosted platform. The binding constraint is not model accurac… (voir plus)y but infrastructure to route studies, display results, capture feedback, and audit what runs. At one center, angiography models completed 515 of 607 jobs (84.8%); failures reflected absent diagnostic views, and 78.1% of 638 clinician ratings were positive. Publishing honest readiness levels for every model is itself a governance practice.
The effects of positive allosteric modulators (PAMs) of the GABA A receptor in disorders of consciousness (DoC)—such as the hypnotic zolpi… (voir plus)dem in coma—are frequently reported but remain poorly understood. Although several hypotheses have been proposed to explain their paradoxical effect, biologically plausible computational models that test why only a subset of patients with DoC respond to inhibitory-enhancing medications may clarify this phenomenon and guide therapeutic strategies. We extended a spiking neural network model to mechanistically investigate inhibitory neuron enhancement in DoC. The model featured biologically inspired elements, including a ratio of 80:20 excitatory-to-inhibitory neurons and 12 Hz of spontaneous intrinsic activity. Neurons were spatially organized, and the network performed a trace conditioning task to investigate conscious access. We implemented localized and diffuse lesions affecting excitatory and inhibitory neuron populations, individually and in combination. The positive allosteric modulation of GABA A receptors was simulated by increasing inhibitory synaptic weights. Experiments revealed distinct recovery patterns depending on damage type and lesion distribution. Spontaneous recovery was more impaired by excitatory than inhibitory lesions, with local lesions generally affecting performance more than diffuse ones. Inhibitory postsynaptic potentiation produced dose-dependent recovery after inhibitory lesions, with therapeutic windows varying by lesion type. Our findings suggest that the paradoxical effect of GABA A receptor PAMs arises from the restoration of excitatory–inhibitory balance when inhibitory networks are moderately disrupted. This computational framework offers a testable account in which insufficient inhibitory regulation represents a common pathway underlying DoC cases responsive to GABA A receptor PAMs across diverse etiologies.
2026-09-21
Proceedings of the National Academy of Sciences (publié)
Language models can be adapted by changing the computations applied to individual tokens. Quantum circuits offer one such approach, but eval… (voir plus)uating wider circuits inside a large model can be computationally demanding. Here we introduce HyperQ, which adds token-conditioned quantum residual branches to a frozen masked-diffusion language model. A quantum residual branch is a module in each transformer block that reads a token's hidden state, emits the coordinates of that token's circuit, executes it, and adds the measured values back through a residual connection. The backbone remains frozen, and only the added branches are trained. Within each branch, a lightweight circuit hypernetwork emits token-specific rotation angles, coupling strengths, and measurement axes in a shared sparse circuit structure. The required expectation values have an exact classical expression whose evaluation cost grows linearly with the qubit count, enabling circuits from 16 to 64 qubits to be trained within a 1.1-billion-parameter backbone. Across downstream benchmarks, increasing circuit width raises the average score from 47.65 to 54.30. At 64 qubits, HyperQ exceeds the backbone and its low-rank-adapted counterpart by 4.71 and 3.67 points, respectively. HyperQ is fine-tuned on 20,000 prompt-response pairs, compared with 200,000 for the classical baselines. These findings support token-conditioned circuit emission as a tractable architectural approach to quantum-augmented language modelling.