Publications

Physics-informed cross-coupled information flow modeling for spatiotemporal dynamical systems
Hangyi Yu
Hanqing Zhao
Yu Zhang
Lianlei Lin
Zongwei Zhang
Sheng Gao
Junkai Wang
2026-04-30
Information Sciences (publié)
doi.org
Position: agentic AI orchestration should be Bayes-consistent
Theodore Papamarkou
Pierre Alquier
Matthias Bauer
Wray Buntine
A. Davison
Gintare Karolina Dziugaite
Maurizio Filippone
Andrew Y. K. Foong
Vincent Fortuin
Dimitris Fouskakis
Jes Frellsen
Eyke Hüllermeier
Theofanis Karaletsos
Mohammad Emtiyaz Khan
Nikita Kotelevskii
Salem Lahlou
Yingzhen Li
Fang Liu
Clare Lyle
Thomas Möllenhoff … (voir 10 de plus)
Konstantina Palla
Maxim Panov
Yusuf Sale
Kajetan Schweighofer
Artem Shelmanov
Siddharth Swaroop
Martin Trapp
Willem Waegeman
Andrew Gordon Wilson
Alexey Zaytsev
LLMs excel at predictive tasks and complex reasoning tasks, but many high-value deployments rely on decisions under uncertainty, for example… (voir plus), which tool to call, which expert to consult, or how many resources to invest. While the usefulness and feasibility of Bayesian approaches remain unclear for LLM inference, this position paper argues that the control layer of an agentic AI system (that orchestrates LLMs and tools) is a clear case where Bayesian principles should shine. Bayesian decision theory provides a framework for agentic systems that can help to maintain beliefs over task-relevant latent quantities, to update these beliefs from observed agentic and human-AI interactions, and to choose actions. Making LLMs themselves explicitly Bayesian belief-updating engines remains computationally intensive and conceptually nontrivial as a general modeling target. In contrast, this paper argues that coherent decision-making requires Bayesian principles at the orchestration level of the agentic system, not necessarily the LLM agent parameters. This paper articulates practical properties for Bayesian control that fit modern agentic AI systems and human-AI collaboration, and provides concrete examples and design patterns to illustrate how calibrated beliefs and utility-aware policies can improve agentic AI orchestration.
2026-04-30
arXiv (prépublication)
doi.org
arxiv.org
Profiling the Cell-Type Specific Effects of Psilocybin in Medial Prefrontal Cortexh
Heike Schuler
Delong Zhou
Chloé Savignac
Vedrana Cvetkovska
Yiu-Chung Tse
Juliet Meccia
Danilo Bzdok
Rosemary C. Bagot
2026-04-30
Biological Psychiatry (publié)
doi.org
Refining the construct of direct verbal suggestibility: Evidence for a hybrid dimensional–typological latent structure
Jérémy Brunel
Audrey Vanhaudenhuyse
Julie Delage
Karim Jerbi CoCo Lab
Pierre Rainville
David Ogez
Mathieu Landry
2026-04-30
Personality and Individual Differences (publié)
doi.org
SCEIMA: Social Coordination Evaluation through Integrated Model Analysis
Bavo Van Kerrebroeck
Caroline Palmėr
Guillaume Dumas
Alexander P. Demos
Computational models are increasingly used as interactive partners in studies of human coordination, yet it remains unclear whether observed… (voir plus) differences in human behavior reflect properties of the models themselves, changes in human behavior elicited by such artificial partners, or both. We introduce SCEIMA (Social Coordination Evaluation through Integrated Model Analysis), a two-stage framework designed to disentangle human-specific, model-specific, and interaction-driven contributions to coordination in human–machine interaction paradigms. In the empirical stage, human participants perform a coordination task with both human partners and computational models, establishing reference human–human and human–model interaction patterns. In the analytical stage, the same models are paired with one another and optimized through simulations to reproduce empirical coordination metrics. Comparing human–human, human–model, and simulated model–model interactions reveals whether coordination differences arise from intrinsic model dynamics, from human adaptation to artificial partners, or from their interaction. SCEIMA treats computational models as contrastive instruments whose capacity to elicit and reproduce human behavior can be systematically evaluated. We illustrate the framework with two distinct case-studies, a sensorimotor synchronization task and a conversational turn-taking task, showing how distinct outcome patterns diagnose the sources of coordination differences. By providing a principled methodological framework for evaluating interactive computational models, SCEIMA improves interpretability in human–machine interaction research and informs the design of artificial agents that coordinate with humans more naturally and responsively.
2026-04-30
Computers in Human Behavior Reports (publié)
doi.org
Supplier-Independent Capability Verification of Software and Firmware with AI-Driven ATT&CK Mapping
Mitchell Petingola
Philippe Charland
Steven H. H. Ding
Benjamin C. M. Fung
In defense and other security-critical domains, software, and firmware must perform only their mission-required functions. Hidden or dormant… (voir plus) capabilities, such as unauthorized data transmission, cryptographic routines, or file manipulation, expand the attack surface because they can later be activated by attackers after deployment. The 2020 SolarWinds supply chain compromise demonstrated how trusted updates can be weaponized, with embedded code that remains inactive at first, but later enables movement through secure networks, while still being undetected. Relying solely on supplier attestations is therefore insufficient. Even reputable vendors may unknowingly deliver unnecessary or vulnerable code. To mitigate this, we advocate independent reverse engineering and capability verification to identify and limit what software and firmware are technically able to do. This verification gives commanders and certifiers confidence that the deployed systems remain secure, predictable, and trustworthy.To systematically classify and communicate verified capabilities, the MITRE Adversarial Tactics, Techniques, and Common Knowledge (ATT&CK) framework provides a suitable taxonomy that is widely adopted across defense and intelligence communities as a common reference for adversary techniques. However, its existing behavior-scanning methods are limited: Dynamic analysis (sandboxing) only observes the code paths that happen to execute, missing latent or environment-dependent functions, while static analysis typically recognizes a limited number of ATT&CK techniques, leaving large blind spots. We address this gap by combining language models with Capa-style rule generation to produce fast, binary-level matching rules. This extends Capa, originally built for malware triage, into a broader verification tool that produces detailed, auditable reports. In experiments on 878 malware samples, our method expanded ATT&CK coverage from 148 to 262 techniques, including high-impact behaviors such as credential dumping and persistence via scheduled tasks. This broader mapping enables more robust independent capability verification, reduces compliance risks, and strengthens assurance for mission-critical defense and security-critical systems.
2026-04-30
Conference on Cyber Conflict (publié)
doi.org
The ethical impasse of current consciousness science
Vincent Taschereau-Dumouchel
Jun Seo Hwang
Hakwan Lau
Joseph E. LeDoux
2026-04-30
Neuron (publié)
doi.org
Training a neural network to rapidly identify candidate gravitational-wave events in the lower mass gap
Nayyer Raza
Man Leong Chan
Daryl Haggard
Ashish Mahabal
Jess McIver
Audrey Durand
Alexandre Larouche
Hadi Moazen
The physics governing the boundary between the most massive neutron stars (NSs) and the least massive black holes (BHs) is currently uncerta… (voir plus)in, but could potentially be constrained with new observations. While NSs have been observed with masses up to
2026-04-30
arXiv (prépublication)
doi.org
arxiv.org
Adaptive Batch Sizes Using Non-Euclidean Gradient Noise Scales for Stochastic Sign and Spectral Descent
Hiroki Naganuma
Shagun Gupta
Youssef Briki
Ioannis Mitliagkas
Irina Rish
Parameswaran Raman
Hao-Jun Michael Shi
To maximize hardware utilization, modern machine learning systems typically employ large constant or manually tuned batch size schedules, re… (voir plus)lying on heuristics that are brittle and costly to tune. Existing adaptive strategies based on gradient noise scale (GNS) offer a principled alternative. However, their assumption of SGD's Euclidean geometry creates a fundamental mismatch with popular optimizers based on generalized norms, such as signSGD / Signum (
2026-04-29
International Conference on Machine Learning (accepté)
doi.org
openreview.net
AES: Curing Optimizer Blindness in Long-Tailed Recognition via State-Aware Correction
Fanfu Wang
Jiachang Zhan
Zhiheng Gong
Pengkun Wang
Yang Wang
Long-tailed recognition fundamentally suffers from optimizer blindness where the optimization process mistakenly conflates the magnitude of … (voir plus)gradient accumulation with the scarcity of semantic information. Existing strategies relying on static frequency-based priors fail to correct this bias and result in state blindness regarding supervision and micro-level blindness regarding parameter updates. To address these limitations, we propose the AES framework to establish a dynamic and state-aware correction system across the entire learning lifecycle. We specifically introduce Adaptive Residual Supervision loss to act as a real-time reality check for supervision completeness via precision shielding. We also propose Entropy-aware PCGrad to resolve parameter-level conflicts by quantifying task specificity through gradient entropy. Additionally, we devise Sample-level Conflict Arbitrated Fusion to serve as a dynamic inference arbiter that routes predictions based on instance difficulty. Extensive experiments on CIFAR-100-LT, ImageNet-LT, and iNaturalist 2018 demonstrate that our method consistently achieves state-of-the-art performance by effectively balancing head-class stability and tail-class discrimination. Code is available at [here](https://github.com/wangff0101/AES)
2026-04-29
International Conference on Machine Learning (accepté)
openreview.net
openreview.net
Ariadne's Thread of LipSync: Unraveling Forgeries via Inconsistency between Lip Motions and Head Poses
Tianyi She
Jiawei Liu
Weifeng Liu
Hanqing Zhao
Weiming Zhang
Kejiang Chen
Recent advances in LipSync generation technology have led to the creation of highly realistic videos, posing severe societal risks. However,… (voir plus) existing defense strategies struggle against LipSync forgeries, as state-of-the-art generative models not only optimize for the lip synchronization but also significantly eliminate visual artifacts, resulting in the lack of key detection signals. Inspired by the inherent biological coupling between lip movements and head poses in natural speech, we observe that generative models fundamentally disrupt this global coordination when optimizing for local lip motion. In this paper, we propose LipDA, a novel framework for joint LipSync Detection and Attribution, which takes advantage of the inconsistency between head and lip. For detection, the framework learns to quantify this discrepancy by contrasting lip and pose features from authentic versus forged videos. For attribution, our method is designed to capture the unique temporal dynamics and audio-visual synchronization patterns that act as generative fingerprints, enabling source tracing. To validate our approach, we conduct extensive experiments on two challenging LipSync benchmarks as well as on our own proposed large-scale and multi-generator dataset, LipSyncBench-A. LipDA achieves over 97% AUC in detection and 97.5% accuracy in model attribution, significantly outperforming existing methods.
2026-04-29
International Conference on Machine Learning (accepté)
openreview.net
openreview.net
Balancing Plasticity and Stability with Fast and Slow Successor Features
Raymond Chua
Doina Precup
Blake Richards
A hallmark of intelligence is the ability to adapt in non-stationary environments, yet deep Reinforcement Learning (RL) agents often struggl… (voir plus)e in such settings. Prior studies introduce non-stationarity through abrupt shifts in features or dynamics, whereas real-world environments often evolve gradually through continual drift. This distinction has important implications for the ``stability-plasticity dilemma'' in RL, as abrupt task changes may demand more plasticity than naturalistic settings. To address this, we modify existing 3D Miniworld and MuJoCo environments to incorporate naturalistic, continual non-stationarity, and use them to examine how stability and adaptation affect performance under continuous environmental change. We find that methods favoring stability, such as synaptic consolidation, outperform approaches focused on plasticity, such as parameters resetting. Motivated by this result, and prior evidence that Successor Features (SFs) reduce interference, we investigate whether SFs are better consolidation targets than Q-values. Across both environments, applying neuro-inspired synaptic consolidation to SFs yields superior performance on continually changing settings. Moreover, consolidation is most effective when SFs are stabilized across multiple timescales, which capture complementary aspects of gradual environmental change. Together, these results suggest that stability is more critical in continual learning when changes are gradual, and that multi-timescale consolidation of predictive representations is an effective approach.
2026-04-29
International Conference on Machine Learning (accepté)
doi.org
openreview.net