Publications

Surrogate Loss Gradient Descent in Variational Inequalities with Linear Approximation
The use of surrogate losses to approximate principled updates in prediction space has been instrumental in the success of scaling function a… (voir plus)pproximation in reinforcement learning. These methods are motivated by approximating an update in prediction space with an optimization inner-loop that takes a fixed number of gradient steps on a surrogate loss. Despite its prevalence, there is a disconnect between existing theory and algorithms used in practice with respect to the number of gradient steps. In this paper, we focus on the linear approximation case to derive sharp guarantees when parameters are updated by a fixed number of gradient steps on a surrogate loss. It provides a unified approach that naturally connects the mean-path analysis of TD and LSPE, which correspond to both extremes from one step to an infinite number of inner-steps, and have been previously analysed with different techniques. Our analysis, however, gives a precise insight into the compute-optimal number of gradient steps
The AI Neuroscientist: An Interactive Agentic Interface for Neuroimaging Analysis
Aakash Patel
Panayiotis Ketonis
Shreya Saxena
J. Adam Noah
David van Dijk
Analyzing neuroimaging data requires specialized coding and statistical expertise, which limits accessibility for researchers without comput… (voir plus)ational backgrounds. We present the AI Neuroscientist, a language agent for interactive data exploration. The system integrates a large language model (LLM) with a neuroimaging toolset to perform quality control, modeling, and visualization. This allows researchers to query data quality and specify analysis parameters directly in natural language, providing a transparent and interactive alternative to conventional scripted pipelines for small-scale data exploration. We demonstrate these capabilities using functional near-infrared spectroscopy (fNIRS) data, and evaluate the agent on a custom fNIRS benchmarking suite against general-purpose LLM agents with code sandboxes. Future extensions will generalize the architecture to additional modalities, including functional magnetic resonance imaging (fMRI) data, and expand the benchmarking suite to additional fNIRS tasks.
Validating gene-brain-behavior associations in autism: A replication study
Dongyue Zhou
Ting Mei
Marianne Oldehinkel
Maarten Mennes
Natalie J. Forde
Flavio Dell’Acqua
Tobias Banaschewski
Christine Ecker
Rosemary Holt
Thomas Bourgeron
Tony Charman
Eva Loth
Declan Murphy
Yunge Zhang
Huanjie Li
Fengyu Cong
Jan K. Buitelaar
Christian F. Beckmann
Jumana Ahmad
Sara Ambrosino … (voir 69 de plus)
Bonnie Auyeung B
Tobias Banaschewski
Simon Baron‐Cohen
Sarah Baumeister
Christian F. Beckmann
Sven Bölte
Thomas Bourgeron
Carsten Bours
Michael John Brammer
Daniel Brandeis
Claudia Brogna
Yvette G. E. de Bruijn
Jan K. Buitelaar
Bhismadev Chakrabarti
Tony Charman
Ineke Cornelissen
Daisy Crawley
Flavio Dell’Acqua
Sarah Durston
Christine Ecker
Jessica Faulkner
Vincent Frouin
Pilar Garcés
David Goyard
Lindsay M. Ham
Hannah Hayward
Joerg F. Hipp
Rosemary Holt
Mark H. Johnson
Emily J. H. Jones
Prantik Kundu
Meng‐Chuan Lai
Xavier Liogier D’ardhuy
Michael Vincent Lombardo
Eva Loth
David J. Lythgoe
René C.W. Mandl
Andre Marquand
Luke Mason
Maarten Mennes
Andreas Meyer‐Lindenberg
Carolin Moessnang
Nico Mueller
Declan Murphy
Bethany Oakley
Laurence O’Dwyer
Marianne Oldehinkel
Bob Oranje
Gahan Pandina
Antonio M. Persico
Jack Price
Annika Rausch
Barbara Ruggeri
Amber N. V. Ruigrok
Jessica Sabet
Roberto Sacco
Antonia San José Cáceres
Emily A. Simonoff
Will P. J. M. Spooren
Julian Tillmann
Roberto Toro
Heike Tost
Jack Waldman
Steve C.R. Williams
Caroline Wooldridge
Iva Ilioska
Ting Mei
Marcel P. Zwiers
BACKGROUND: Autism is characterized by heterogeneity in clinical characteristics, neuroimaging and genetic profiles. To elucidate the multi-… (voir plus)level heterogeneity, Buch et al. (2023) identified biologically meaningful subgroups in Autism Brain Imaging Data Exchange (ABIDE). However, ABIDE contains non-biological sources of variability due to retrospectively aligned site protocols. To separate biological from site-related variability and test the robustness of these subgroups, we assessed how well their findings replicate in a deeply phenotyped, prospectively collected autism dataset (EU-AIMS LEAP). METHODS: We included 270 autistic (7.5-30.3 years; 73% male) and 216 non-autistic participants (6.9-31 years; 63.4% male) from LEAP. Regularized canonical correlation analysis (RCCA) was conducted between functional connectivity and three clinical measures (verbal IQ [VIQ], and Autism Diagnostic Observation Schedule-2 [ADOS-2] social affect [SA] and restricted/repetitive behavior [RRB]). Next, subgroups identified by hierarchical clustering based on canonical variates were characterized by behavior, functional connectivity patterns and gene enrichment analyses. RESULTS: Three of four autistic subgroups showed behavioral profiles consistent with the original study. Three subgroups shared stronger thalamocortical connectivity, while all four displayed distinct connectivity involving default mode network (DMN), salience (SN), visual and somatomotor networks. These subgroup-specific connectivity patterns were further linked to autism-associated and attention deficit hyperactivity disorder (ADHD)-related gene sets. Across preprocessing pipelines, subgroup assignments were partially preserved. Subgroup2 (globally affected behaviors) and Subgroup4 (less social difficulty, but more restricted behavior) retained more consistent behavioral profiles and connectivity alterations, whereas Subgroup1 and Subgroup3 were more sensitive to the precise preprocessing steps. CONCLUSIONS: These four subgroups showed partly consistent behavioral profiles, connectivity alterations and genetic patterns, extending our understanding of the multi-level heterogeneity of autism.
Ambiguity-Aware Evaluation for Automated Fact-Checking
Sukanya Krishna
Maimuna S. Majumder
Finale Doshi-Velez
Automated fact-checking systems are evaluated almost entirely by whether their predicted verdict matches a single gold label, an approach th… (voir plus)at assumes every claim has one stable interpretation and that the available evidence is always sufficient to resolve it. In practice, claims are frequently underspecified in time, place, or meaning, and retrieved evidence is often incomplete or contradictory, so this approach cannot distinguish a system that reasoned well on a genuinely ambiguous input from one that reasoned poorly but matched the label anyway. We introduce a framework that separates ambiguity in the claim and evidence from failures in the system's own reasoning and grounding, and conditions accuracy and response appropriateness on this ambiguity profile. Applied to a retrieval-based fact-checking pipeline across five benchmarks, plus a human pilot study, we find that ambiguity is pervasive and dominated by missing timeframes, that most correct verdicts on several benchmarks are backed by reasoning that would not hold up under scrutiny, and that systems grow more confident, not less, when their evidence conflicts.
AnchorPrompt: Self-Distilled Soft Prompts for Robust Audio-Language Models
Large audio-language models (LALMs) are sensitive to input perturbations, such as noise, waveform corruption, and adversarial injections. We… (voir plus) propose AnchorPrompt, an efficient adaptation method that keeps the model frozen and learns a single block of prompt vectors inserted at the decoder input, between the audio and question embeddings. We train these vectors through self-distillation over diverse audio and text perturbations. To improve answer consistency and mitigate hallucination, we use the model's prediction on the clean recording as the target for answerable inputs, and assign a refusal target when the audio lacks sufficient evidence to answer. Furthermore, AnchorPrompt is perturbation-agnostic at inference, requiring no prior detection of perturbations and enabling zero-shot transfer to unseen distortions. We evaluate three LALMs across three benchmarks and show that AnchorPrompt improves answer consistency in most tested conditions. Clean accuracy improves in six of nine model-benchmark pairs, with minimal impact on the remainder of 1.2% at most. Crucially, AnchorPrompt reduces hallucinations under severe audio corruption while keeping false refusals on clean audio rare. Finally, these consistency gains transfer to unseen perturbations, such as choice permutations and reverberation.
Authentic participatory research with older adults for cognitive health (APROACH) in rural Botswana and urban Montreal: protocol for a pilot randomised controlled trial
Neil Andersson
Anne Cockcroft
Machelle Wilchesky
Ivan Sarmiento
Ogopoleng Batisi
Mallar Chakravarty
Maiya Geddes
Franque Grimard
Vladimir Hachinski
Claudia Mitchell
Kathy Pichora-Fuller
Samira Abbasgholizadeh Rahimi
Kim Sawchuk
Walter J. Wittich
Lehana Thabane
INTRODUCTION: Increasing numbers of older adults live with dementia especially in low- and middle-income countries. Most prevention relies o… (voir plus)n controlling conditions that increase the risk of dementia, such as hypertension, diabetes and sight and hearing loss. Recent pharmacological treatments have limited efficacy and are prohibitively expensive. Social isolation increases the risk of dementia and evidence suggests interventions to increase social engagement can reduce the risk possibly by supporting executive function. Three key features of authentic participatory research - shared governance, use of evidence and innovation - align with elements of executive function. Our pilot tests the feasibility and acceptability of participatory dementia prevention in Montreal, Canada and rural Botswana. METHODS AND ANALYSIS: This pilot randomised controlled trial will recruit women and men aged 65 years and older who are socially isolated (not in paid or voluntary employment, not married) in urban Montreal and two rural communities in Botswana. In Botswana, we will randomly allocate participants in one community to receive the intervention immediately while the control community will receive the intervention after 18 months. In Montreal, individual participants will be randomly allocated to immediate or delayed intervention groups before or after their grouping by shared concern. We aim to recruit at least 100 participants in Botswana and 100 in Montreal. The intervention has six steps: individuals identify priority concerns; participants join groups with shared concerns; groups discuss the concern and potential solutions; groups share the concern and possible solutions with other stakeholders; implementation of solutions; and self-evaluation of the process. We will measure executive function using the Frontal Assessment Battery, and apply a questionnaire to measure integral brain health, physical health and social isolation at baseline, at 18 months and at 33 months. We will analyse narratives of change at the end of the intervention periods to understand participants' experience of the intervention. ETHICS AND DISSEMINATION: The McGill University Faculty of Medicine and Health Sciences Institutional Review Board approved the entire project on 18 September 2024, Study number: A06-M27-24A. The Psychosocial Research Committee (REC) of CIUSSS West-Central Montreal Research Ethics Board (REB) approved recruitment of clinic-related participants in Montreal on 3 March 2025, Project 2025-4112. The Health Research and Development Division of the Ministry of Health, Botswana, approved the study in Botswana on 24 June 2024, Reference number: HPRD 6/14/1.In Botswana, we will disseminate the findings in community meetings in the two pilot communities and with government and non-government stakeholders convened by the Ministry of Health and the Social Development Department of the Ministry of Local Government and Traditional Affairs. In Montreal, we will disseminate the findings in research meetings and in other relevant fora. We will give oral and poster presentations about the study in relevant local, national and international academic conferences and meetings and publish papers about the findings in peer-reviewed, indexed academic journals. TRIAL REGISTRATION NUMBER: ISRCTN14273485.
Authentic participatory research with older adults for cognitive health (APROACH) in rural Botswana and urban Montreal: protocol for a pilot randomised controlled trial
Neil Andersson
Anne Cockcroft
Machelle Wilchesky
Ivan Sarmiento
Ogopoleng Batisi
Mallar Chakravarty
Maiya Geddes
Franque Grimard
Vladimir Hachinski
Claudia Mitchell
Kathy Pichora-Fuller
Samira Abbasgholizadeh Rahimi
Kim Sawchuk
Walter J. Wittich
Lehana Thabane
INTRODUCTION: Increasing numbers of older adults live with dementia especially in low- and middle-income countries. Most prevention relies o… (voir plus)n controlling conditions that increase the risk of dementia, such as hypertension, diabetes and sight and hearing loss. Recent pharmacological treatments have limited efficacy and are prohibitively expensive. Social isolation increases the risk of dementia and evidence suggests interventions to increase social engagement can reduce the risk possibly by supporting executive function. Three key features of authentic participatory research - shared governance, use of evidence and innovation - align with elements of executive function. Our pilot tests the feasibility and acceptability of participatory dementia prevention in Montreal, Canada and rural Botswana. METHODS AND ANALYSIS: This pilot randomised controlled trial will recruit women and men aged 65 years and older who are socially isolated (not in paid or voluntary employment, not married) in urban Montreal and two rural communities in Botswana. In Botswana, we will randomly allocate participants in one community to receive the intervention immediately while the control community will receive the intervention after 18 months. In Montreal, individual participants will be randomly allocated to immediate or delayed intervention groups before or after their grouping by shared concern. We aim to recruit at least 100 participants in Botswana and 100 in Montreal. The intervention has six steps: individuals identify priority concerns; participants join groups with shared concerns; groups discuss the concern and potential solutions; groups share the concern and possible solutions with other stakeholders; implementation of solutions; and self-evaluation of the process. We will measure executive function using the Frontal Assessment Battery, and apply a questionnaire to measure integral brain health, physical health and social isolation at baseline, at 18 months and at 33 months. We will analyse narratives of change at the end of the intervention periods to understand participants' experience of the intervention. ETHICS AND DISSEMINATION: The McGill University Faculty of Medicine and Health Sciences Institutional Review Board approved the entire project on 18 September 2024, Study number: A06-M27-24A. The Psychosocial Research Committee (REC) of CIUSSS West-Central Montreal Research Ethics Board (REB) approved recruitment of clinic-related participants in Montreal on 3 March 2025, Project 2025-4112. The Health Research and Development Division of the Ministry of Health, Botswana, approved the study in Botswana on 24 June 2024, Reference number: HPRD 6/14/1.In Botswana, we will disseminate the findings in community meetings in the two pilot communities and with government and non-government stakeholders convened by the Ministry of Health and the Social Development Department of the Ministry of Local Government and Traditional Affairs. In Montreal, we will disseminate the findings in research meetings and in other relevant fora. We will give oral and poster presentations about the study in relevant local, national and international academic conferences and meetings and publish papers about the findings in peer-reviewed, indexed academic journals. TRIAL REGISTRATION NUMBER: ISRCTN14273485.
Beyond Fair Answers: Learning Fair and Unlearning Biased Reasoning in LLMs
Large Language Models (LLMs) have demonstrated strong reasoning capabilities on tasks with well-defined solutions, such as mathematical and … (voir plus)logical reasoning. However, reasoning capability does not necessarily imply unbiased reasoning. In socially ambiguous contexts, where the available evidence is insufficient to support one conclusion over another, models may rely on demographic stereotypes or unsupported social assumptions. Recent work shows that chain-of-thought (CoT) reasoning can contain social stereotypes and that biased reasoning steps are associated with stereotype expression and incorrect predictions. This motivates evaluating fairness in the reasoning process, not only in the final answer. We therefore ask: Can we train LLMs not only to produce fair answers, but to reason fairly? We study this question using a fairness-oriented adaptation of BBQ, where ambiguous cases test whether demographic alternatives receive comparable probability when evidence is insufficient, while disambiguated cases test evidence-grounded accuracy. Following reasoning-trace fine-tuning approaches such as STaR and ReGiFT, we fine-tune instruction-tuned LLMs on reasoning traces generated by stronger models and compare base and fine-tuned models under direct and CoT inference. We evaluate both final decisions and generated reasoning traces, and test generalization on SocioEconomicQA and CrowS-Pairs. Building on this analysis, we propose a reasoning-level fairness framework that jointly learns fair reasoning and unlearns biased reasoning. Evidence-grounded traces form the positive learning set, while traces containing unsupported demographic or stereotype-based inferences form the forget set. Our objective combines supervised fine-tuning on fair traces, a negative objective on biased traces, and KL regularization on unrelated reasoning data to preserve general reasoning ability. We evaluate whether this process improves selection of the fair answer and, more importantly, reduces the probability gap between gendered options when the context provides no evidence favoring either one. We separately measure accuracy in disambiguated contexts to ensure that the model still uses available evidence correctly. At the reasoning level, we evaluate stereotype-dependent inferences and the relative likelihood of fair versus biased reasoning traces. We compare against ReGiFT, standard CoT, and instruction-based reasoning baselines, and evaluate generalization across additional bias benchmarks while monitoring general reasoning performance for capability loss. This work moves fairness intervention from the final decision to the reasoning process that produces it. By combining fair-reasoning supervision, biased-reasoning unlearning, and capability preservation, we investigate whether LLMs can learn not only to give fair answers, but to reason without relying on unsupported demographic stereotypes.
Gated Knowledge-Guided Prototype Transfer for Zero-Support Medical Vision-Language Adaptation
Sumaiya Fazal
Rizwan Qureshi
Realistic few-shot adaptation of medical vision-language models (VLMs) can leave low-prevalence classes without any labeled support. The tra… (voir plus)ining-free ss-text+ solver remains operational in this setting, but its zero-support prototypes receive no visual adaptation and fall back to a completely textual class prototype. We introduce gated Knowledge-Guided Prototype Transfer gated KGPT, a pre-repulsion module that transfers supported-class visual residuals to missing classes using VLM similarity and an ontology-based clinical distance. A validation-free evidence gate scales transfer based on semantic relevance and residual agreement, thereby limiting unsupported extrapolation. In 50 paired realistic one-shot trials on MESSIDOR with FLAIR features, severe and proliferative diabetic retinopathy were absent from every support set. Relative to ss-text+, gated KGPT improved missing-class recall by 5.23\%, average class accuracy by 1.37\%, and the missing/supported recall harmonic mean by 1.83\%. It reduced negative-transfer frequency from 46 to 48\% for ungated transfer to 2\%, while leads to 1.20\% supported-class recall reduction and a 0.45\% top-1 reduction. These results identify that evidence-based gating provides a safety mechanism for the knowledge source.
Grid Cells as Spectral Successor Features: A Reusable Basis for Continual Reinforcement Learning
A reinforcement-learning agent moving between tasks in a fixed environment faces a mismatch: the reward changes, but the geometry does not. … (voir plus)Methods that learn a representation and a value function jointly relearn both at every task switch, which is slow and prone to forgetting. We take the identification of grid cells with eigenvectors of the successor representation as a design principle, and build the feature basis from the low-frequency eigenvectors of the symmetrised random-walk graph Laplacian, defined over state-action pairs rather than states so that the features encode what an action does. Because this basis diagonalises the transition operator, the successor-feature transform is exactly diagonal with closed-form entries fixed by the Laplacian eigenvalues and the discount. No temporal-difference learning is needed for it, and neither a successor-feature network nor a replay buffer is required: the only learned quantity is a linear reward map, re-estimated per task. On two goal locations in a two-room maze sharing an identical transition structure, the resulting agent holds near-ceiling return across every task switch, where a successor-feature baseline that learns its features collapses at each change. The cost is a slightly lower ceiling imposed by the linear parameterisation. The basis here is computed from the true transition graph, so this arm is an oracle upper bound rather than a method competing on equal information; learning the basis online is in progress.
LatentHarness: Learning Latent Actions for Memory and Reasoning via Counterfactual Policy Distillation
Long-context reasoning faces two complementary bottlenecks: retaining evidence across long inputs and sustaining computation across many rea… (voir plus)soning steps. Existing approaches largely address them separately, with external memory extending access to distant evidence and latent reasoning compressing multi-step computation. We introduce LatentHarness, which unifies memory access and latent reasoning as sequential latent action selection. At each internal step, the model chooses THINK for further computation, RECALL from a fast-weight memory of input evidence and intermediate reasoning states, or EXIT to emit the next token. We train this policy with counterfactual policy distillation, which branches every action for one step and scores its effect on the emitted token. These gains teach the policy when memory is more useful than further reasoning, while gradients through counterfactual recall teach which intermediate states should be retained in memory for future use. Across six general and long-context reasoning benchmarks, LatentHarness at 1.4B improves on the strongest baselines by 2.8% and 10.0% relative, respectively, and runs 5.9x faster than the strongest long-context baseline.
Less is more: error-distance scaling relation for data-efficient kilometer-scale downscaling of extreme heat
Ahmed Marey
Henry Lu
Abhishek Gaur
Sherif Goubran
Malek Aloui
Theodore Potsis
Liangzhu Leon Wang
Extreme heat is where urban adaptation needs kilometer-scale data the most, but the simulations training a downscaler can cost more than the… (voir plus)y save, and how much is needed has not been identified. We measured it with CASPER, a U-Net with a structure-preserving loss downscaling 32 km reanalysis to 1 km temperature, humidity and wind, across 24 configurations of one to eight months. Held-out error grows linearly with climatological distance to the training data, RMSE = 0.83 + 2.95 d, explaining 90% of its variance against 7% for volume and predicting unseen months in advance. On held-out extreme summer weeks CASPER preserves the fine-scale structure and cross-variable physics that matched-budget baselines degrade, and matches station observations during documented heat waves to within 1.8 K. Transfer to a new region degrades geographically; 11 days of local simulation cuts Vancouver's held-out error from 3.8 to 1.3 K. Training periods should span the target climate: the same accuracy for four times less simulation, putting kilometer-scale downscaling of extreme heat within reach of groups without large computing facilities.