Publications

Revisiting the Learning Objectives of Vision-Language Reward Models
Simon Roy
Samuel Barbeau
Giovanni Beltrame
Christian Desrosiers
Nicolas Thome
Learning generalizable reward functions is a core challenge in embodied intelligence. Recent work leverages contrastive vision language mode… (see more)ls (VLMs) to obtain dense, domain-agnostic rewards without human supervision. These methods adapt VLMs into reward models through increasingly complex learning objectives, yet meaningful comparison remains difficult due to differences in training data, architectures, and evaluation settings. In this work, we isolate the impact of the learning objective by evaluating recent VLM-based reward models under a unified framework with identical backbones, finetuning data, and evaluation environments. Using Meta-World tasks, we assess modeling accuracy by measuring consistency with ground truth reward and correlation with expert progress. Remarkably, we show that a simple triplet loss outperforms state-of-the-art methods, suggesting that much of the improvements in recent approaches could be attributed to differences in data and architectures.
2025-12-19
ArXiv (preprint)
doi.org
arxiv.org
A Message from AI Research Leaders: Join Us in Supporting OpenReview
Joelle Pineau
Doina Precup
Kyunghyun Cho
Andrew Y. Ng
Yoshua Bengio
Ruslan Salakhutdinov
Fernando Pereira
2025-12-17
OpenReview (unknown)
openreview.net
openreview.net
Most German Speakers Ignore the Cue That Best Predicts Plural Class
Kate McCurdy
Timothy J. O'Donnell
Adam Lopez
Sharon Goldwater
Researchers generally assume that speakers use the linguistic information available to them. For instance, if one grammatical category robus… (see more)tly predicts another grammatical category, we expect speakers to reproduce this conditional relationship during language production. Here, we investigate this assumption for grammatical gender in German. Gender is the single cue which most strongly predicts the plural class of existing German nouns, but behavioral studies with novel nouns have found mixed results regarding the role of gender in plural generalization. Across three experiments, we examine how individual German speakers use grammatical gender when producing plural forms of novel nouns. We find that most speakers effectively ignore gender during plural class production, even under experimental manipulations that encourage them to attend to this cue. These results point toward an underexplored direction in cognitive science: accounting for the linguistic information that speakers do not use.
2025-12-17
Open Mind (published)
doi.org
Imitation Game: Reproducing Deep Learning Bugs Leveraging an Intelligent Agent
Mehil B. Shah
Mohammad Masudur Rahman
Foutse Khomh
Despite their wide adoption in various domains (e.g., healthcare, finance, software engineering), Deep Learning (DL)-based applications suff… (see more)er from many bugs, failures, and vulnerabilities. Reproducing these bugs is essential for their resolution, but it is extremely challenging due to the inherent nondeterminism of DL models and their tight coupling with hardware and software environments. According to recent studies, only about 3% of DL bugs can be reliably reproduced using manual approaches. To address these challenges, we present RepGen, a novel, automated, and intelligent approach for reproducing deep learning bugs. RepGen constructs a learning-enhanced context from a project, develops a comprehensive plan for bug reproduction, employs an iterative generate-validate-refine mechanism, and thus generates such code using an LLM that reproduces the bug at hand. We evaluate RepGen on 106 real-world deep learning bugs and achieve a reproduction rate of 80.19%, a 19.81% improvement over the state-of-the-art measure. A developer study involving 27 participants shows that RepGen improves the success rate of DL bug reproduction by 23.35%, reduces the time to reproduce by 56.8%, and lowers participants'cognitive load.
2025-12-16
ArXiv (preprint)
doi.org
arxiv.org
Effect of Document Packing on the Latent Multi-Hop Reasoning Capabilities of Large Language Models
Gabriele Prato
Shagun Sodhani
Alessandro Sordoni
A. Chandar
The standard practice for training large language models involves packing multiple documents together to optimize computational efficiency. … (see more)However, the impact of this process on the models' capabilities remains largely unexplored. To address this gap, we investigate how different document-packing strategies influence the latent multi-hop reasoning abilities of LLMs. Our findings indicate that packing can improve model performance compared to training on individual documents, at the expense of more compute. To further understand the underlying mechanisms, we conduct an ablation study, identifying key factors that explain the advantages of packing. Ultimately, our research deepens the understanding of LLM training dynamics and provides practical insights for optimizing model development.
2025-12-15
ArXiv (preprint)
doi.org
arxiv.org
Generative epigenetic landscapes map the topology and topography of cell fates.
Victoria Mochulska
Paul François
Epigenetic landscapes were proposed by Waddington as the central concept to describe cell fate dynamics in a locally low-dimensional space. … (see more)In modern landscape models, attractors represent cell types, and stochastic jumps and bifurcations drive cellular decisions, allowing for quantitative and predictive descriptions. However, given a biological problem of interest, we still lack tools to infer and build possible Waddington landscapes systematically. In this study, we propose a generative model for deriving epigenetic landscapes compatible with data. To build the landscapes, we combine gradient and rotational vector fields composed of locally weighted elements that encode ‘valleys’ of the Waddington landscape, resulting in interpretable models. We optimize landscapes through computational evolution and illustrate our approach with two developmental examples: metazoan segmentation and neuromesoderm differentiation. In both cases, we obtain ensembles of solutions that reveal both known and novel landscapes in terms of topology and bifurcations. Conversely, topographic features appear strongly constrained by dynamical data, which suggests that our approach can generically derive interpretable and predictive epigenetic landscapes.
2025-12-15
PubMed (unknown)
doi.org
An Empirical Study of Self-Admitted Technical Debt in Machine Learning Software
Aaditya Bhatia
Foutse Khomh
Bram Adams
Ahmed E. Hassan
The emergence of open-source ML libraries such as TensorFlow and Google Auto ML has enabled developers to harness state-of-the-art ML algori… (see more)thms with minimal overhead. However, during this accelerated ML development process, said developers may often make sub-optimal design and implementation decisions, leading to the introduction of technical debt that, if not addressed promptly, can have a significant impact on the quality of the ML-based software. Developers frequently acknowledge these sub-optimal design and development choices through code comments during software development. These comments, which often highlight areas requiring additional work or refinement in the future, are known as self-admitted technical debt (SATD). This paper aims to investigate SATD in ML code by analyzing 318 open-source ML projects across five domains, along with 318 non-ML projects. We detected SATD in source code comments throughout the different project snapshots, conducted a manual analysis of the identified SATD sample to comprehend the nature of technical debt in the ML code, and performed a survival analysis of the SATD to understand the evolution of such debts. We observed: i) Machine learning projects have a median percentage of SATD that is twice the median percentage of SATD in non-machine learning projects. ii) ML pipeline components for data preprocessing and model generation logic are more susceptible to debt than model validation and deployment components. iii) SATDs appear in ML projects earlier in the development process compared to non-ML projects. iv) Long-lasting SATDs are typically introduced during extensive code changes that span multiple files exhibiting low complexity.
2025-12-14
ACM Transactions on Software Engineering and Methodology (published)
doi.org
arxiv.org
A French Canadian adaptation and validation of the Vividness of Visual Imagery Questionnaire and Plymouth Sensory Imagery Questionnaire.
Catherine Landry
Laurence Lessard
Jeffrey Saint-Louis
Frédéric Gosselin
Ian Charest
Vincent Taschereau-Dumouchel
Guillaume T. Vallet
Mental imagery plays a central role in various cognitive processes and is increasingly investigated in cognitive science. Yet standardized t… (see more)ools for its assessment in French-speaking populations remain scarce. This study examined the psychometric properties of two widely used self-report instruments of mental imagery within the French Canadian population: the Vividness of Visual Imagery Questionnaire (VVIQ) and the Plymouth Sensory Imagery Questionnaire (Psi-Q)-here termed VVIQ-Québec (QC) and Psiq-QC. A total of 328 adults completed the VVIQ-QC and Psiq-QC, with a randomly selected subsample (n = 73) repeating the assessment 1 month later. Exploratory factor analysis of the VVIQ-QC (eyes-open) revealed, as in the original VVIQ, distinct factors corresponding to each prompt cluster (i.e., relative, sunrise, landscape, and storefront). The Psiq-QC yielded a six-factor solution after excluding the Body modality, diverging from the original seven-factor model. Both instruments showed strong internal consistency, temporal stability, and convergent validity. No significant effects of age, sex, or education were observed on imagery scores. These findings provide the first validated French Canadian versions of the VVIQ and Psi-Q, available via the Open Science Framework at https://osf.io/wuhja, offering reliable tools for both research and clinical applications. (PsycInfo Database Record (c) 2025 APA, all rights reserved).
2025-12-14
Canadian Journal of Experimental Psychology / Revue canadienne de psychologie expérimentale (published)
doi.org
QMon: Monitoring the Execution of Quantum Circuits with Mid-Circuit Measurement and Reset
Ning Ma
Jianjun Zhao
Foutse Khomh
Shaukat Ali
Heng Li
Unlike classical software, where logging and runtime tracing can effectively reveal internal execution status, quantum circuits possess uniq… (see more)ue properties, such as the no-cloning theorem and measurement-induced collapse, that prevent direct observation or duplication of their states. These characteristics make it especially challenging to monitor the execution of quantum circuits, complicating essential tasks such as debugging and runtime monitoring. This paper presents QMON, a practical methodology that leverages mid-circuit measurements and reset operations to monitor the internal states of quantum circuits while preserving their original runtime behavior. QMON enables the instrumentation of monitoring operators at developer-specified locations within the circuit, allowing comparisons between expected and observed quantum-state probabilities at those locations. We evaluated QMON by analyzing its impact on circuit behavior, monitoring coverage, and effectiveness in bug localization. Experimental results involving 154 quantum circuits show that all circuits preserve their intended functionality after instrumentation and that QMON successfully detects and localizes various programming errors. Although monitoring coverage is limited by the need to preserve delicate quantum properties, such as entanglement, QMON effectively detects errors while introducing no or negligible disturbance to the original quantum states. QMON facilitates the development of more robust and reliable quantum software as the field continues to mature.
2025-12-14
ArXiv (preprint)
doi.org
arxiv.org
Sliding Window Recurrences for Sequence Models
Dragos Secrieru
Garyk Brixi
Yoshua Bengio
Taiji Suzuki
Michael Poli
Stefano Massaroli
Multi-hybrid architectures are poised to take over language modeling due to better quality and performance. We introduce a hierarchical deco… (see more)mposition framework for linear recurrences that allows us to develop algorithms aligned with GPU memory hierarchies, yielding Sliding Window Recurrences. We focus specifically on truncating recurrences to hardware-aligned windows which are naturally jagged, limiting costly inter-warp communication. Using SWR, we develop Phalanx layers that serve as drop-in replacements for windowed attention or linear recurrences. In 1B parameter multi-hybrid models, Phalanx achieves over 10-40% speedup across 4K to 32K context length over optimized Transformers while matching perplexity.
2025-12-14
arXiv (preprint)
doi.org
arxiv.org
World Models Can Leverage Human Videos for Dexterous Manipulation
Raktim Gautam Goswami
Amir Bar
David Fan
Tsung-Yen Yang
Gaoyue Zhou
Prashanth Krishnamurthy
Michael G. Rabbat
Farshad Khorrami
Yann Lecun
Dexterous manipulation is challenging because it requires understanding how subtle hand motion influences the environment through contact wi… (see more)th objects. We introduce DexWM, a Dexterous Manipulation World Model that predicts the next latent state of the environment conditioned on past states and dexterous actions. To overcome the scarcity of dexterous manipulation datasets, DexWM is trained on over 900 hours of human and non-dexterous robot videos. To enable fine-grained dexterity, we find that predicting visual features alone is insufficient; therefore, we introduce an auxiliary hand consistency loss that enforces accurate hand configurations. DexWM outperforms prior world models conditioned on text, navigation, and full-body actions, achieving more accurate predictions of future states. DexWM also demonstrates strong zero-shot generalization to unseen manipulation skills when deployed on a Franka Panda arm equipped with an Allegro gripper, outperforming Diffusion Policy by over 50% on average in grasping, placing, and reaching tasks.
2025-12-14
ArXiv (preprint)
doi.org
arxiv.org
Disentangling AGN Feedback and Sloshing in the Perseus Cluster with XRISM: Insights from Simulations
Elena Bellomi
John A. ZuHone
Nhut Truong
Irina Zhuravleva
Rainer Weinberger
Christoph Pfrommer
Congyao Zhang
Annie Heinrich
Mateusz Ruszkowski
Brian McNamara
Julie Hlavacek-Larrondo
M. Gendron-Marsolais
Benjamin Vigneron
High-resolution X-ray spectroscopy with XRISM has revealed complex, non-monotonic velocity dispersion profiles in the Perseus cluster, point… (see more)ing to a complex interplay between at least two physical drivers of motions caused by dynamical processes within the intracluster medium (ICM). To further explore this conclusion, we perform a suite of idealized, controlled simulations targeting the relative roles of merger-induced sloshing and active galactic nucleus (AGN) feedback. Our models systematically isolate and combine these mechanisms to predict observable velocity profiles and X-ray line shapes, providing direct comparison to XRISM and Hitomi data. We find that neither sloshing nor AGN activity alone can reproduce the observed velocity dispersion profile; only their combined action matches the elevated dispersions both at the cluster core and outskirts. Power-spectrum analysis reveals distinct spatial signatures: sloshing generates large-scale coherent motions, while AGN feedback injects turbulence and broadens the velocity spectrum at small scales, especially in the core. By forward-modeling spectral line profiles, we show how these dynamics imprint unique observational signatures on X-ray emission. Our results underscore the necessity of accounting for both large-scale and small-scale drivers of gas motions in the ICM when interpreting high-resolution spectroscopic data, and provide guidance for the analysis of forthcoming XRISM observations.
2025-12-13
arXiv (preprint)
doi.org
arxiv.org