Publications

Mitigating Shortcut Learning with Diffusion Counterfactuals and Diverse Ensembles
Luca Scimeca
Alexander Rubinstein
Damien Teney
Seong Joon Oh
Armand Mihai Nicolicioiu
Yoshua Bengio
Spurious correlations in the data, where multiple cues are predictive of the target labels, often lead to a phenomenon known as shortcut lea… (see more)rning, where a model relies on erroneous, easy-to-learn cues while ignoring reliable ones. In this work, we propose
2025-03-05
ICLR.cc/2025/Workshop/SCSL (published)
openreview.net
openreview.net
Mixed Patch Visible-Infrared Modality Agnostic Object Detection
Heitor Rapela Medeiros
David Latortue
Éric Granger
Marco Pedersoli
In real-world scenarios, using multiple modalities like visible (RGB) and infrared (IR) can greatly improve the performance of a predictive … (see more)task such as object detection (OD). Multimodal learning is a common way to leverage these modalities, where multiple modality-specific encoders and a fusion module are used to improve performance. In this paper, we tackle a different way to employ RGB and IR modalities, where only one modality or the other is observed by a single shared vision encoder. This realistic setting requires a lower memory footprint and is more suitable for applications such as autonomous driving and surveillance, which commonly rely on RGB and IR data. However, when learning a single encoder on multiple modalities, one modality can dominate the other, producing un-even recognition results. This work investigates how to efficiently leverage RGB and IR modalities to train a common transformer-based OD vision encoder while countering the effects of modality imbalance. For this, we introduce a novel training technique to Mix Patches (MiPa)from the two modalities, in conjunction with a patch-wise modality agnostic module, for learning a common representation of both modalities. Our experiments show that MiPa can learn a representation to reach competitive results on traditional RGB/IR benchmarks while only requiring a single modality during inference. Our code is available at: https://github.com/heitorrapela/MiPa.
2025-03-05
2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) (published)
doi.org
arxiv.org
A Realistic Protocol for Evaluation of Weakly Supervised Object Localization
Shakeeb Murtaza
Soufiane Belharbi
Marco Pedersoli
Éric Granger
Weakly Supervised Object Localization (WSOL) allows training deep learning models for classification and localization (LOC) using only globa… (see more)l class-level labels. The absence of bounding box (bbox) supervision during training raises challenges in the literature for hyper-parameter tuning, model selection, and evaluation. WSOL methods rely on a validation set with bbox annotations for model selection, and a test set with bbox annotations for threshold estimation for producing bboxes from localization maps. This approach, however, is not aligned with the WSOL setting as these annotations are typically unavailable in real-world scenarios. Our initial empirical analysis shows a significant decline in LOC performance when model selection and threshold estimation rely solely on class labels and the image itself, respectively, compared to using manual bbox annotations. This highlights the importance of incorporating bbox labels for optimal model performance. In this paper, a new WSOL evaluation protocol is proposed that provides LOC information without the need for manual bbox annotations. In particular, we generated noisy pseudo-boxes from a pretrained off-the-shelf region proposal method such as Selective Search, CLIP, and RPN for model selection. These bboxes are also employed to estimate the threshold from LOC maps, circumventing the need for test-set bbox annotations. Our experiments with several WSOL methods on ILSVRC and CUB datasets show that using the proposed pseudo-bboxes for validation facilitates the model selection and threshold estimation, with LOC performance comparable to those selected using GT bboxes on the validation set and threshold estimation on the test set. It also outperforms models selected using class-level labels, and then dynamically thresholded based solely on LOC maps.
2025-03-05
2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) (published)
doi.org
arxiv.org
Shaping Inductive Bias in Diffusion Models through Frequency-Based Noise Control
Thomas Jiralerspong
Berton Earnshaw
Jason Hartford
Yoshua Bengio
Luca Scimeca
Diffusion Probabilistic Models (DPMs) are powerful generative models that have achieved unparalleled success in a number of generative tasks… (see more). In this work, we aim to build inductive biases into the training and sampling of diffusion models to better accommodate the target distribution of the data to model. For topologically structured data, we devise a frequency-based noising operator to purposefully manipulate, and set, these inductive biases. We first show that appropriate manipulations of the noising forward process can lead DPMs to focus on particular aspects of the distribution to learn. We show that different datasets necessitate different inductive biases, and that appropriate frequency-based noise control induces increased generative performance compared to standard diffusion. Finally, we demonstrate the possibility of ignoring information at particular frequencies while learning. We show this in an image corruption and recovery task, where we train a DPM to recover the original target distribution after severe noise corruption.
2025-03-05
ICLR.cc/2025/Workshop/DeLTa (poster)
doi.org
openreview.net
Solving Bayesian Inverse Problems with Diffusion Priors and Off-Policy RL
Luca Scimeca
Siddarth Venkatraman
Moksh Jain
Minsu Kim
Marcin Sendera
Mohsin Hasan
Luke Rowe
Sarthak Mittal
Pablo Lemos
Emmanuel Bengio
Alexandre Adam
Jarrid Rector-Brooks
Yashar Hezaveh
Laurence Perreault-Levasseur
Yoshua Bengio
Glen Berseth
Nikolay Malkin
This paper presents a practical application of Relative Trajectory Balance (RTB), a recently introduced off-policy reinforcement learning (R… (see more)L) objective that can asymptotically solve Bayesian inverse problems optimally. We extend the original work by using RTB to train conditional diffusion model posteriors from pretrained unconditional priors for challenging linear and non-linear inverse problems in vision, and science. We use the objective alongside techniques such as off-policy backtracking exploration to improve training. Importantly, our results show that existing training-free diffusion posterior methods struggle to perform effective posterior inference in latent space due to inherent biases.
2025-03-05
ICLR.cc/2025/Workshop/DeLTa (poster)
doi.org
openreview.net
Towards personalized healthcare without harm via bias modulation
Frank Ngaha
Patrik Joslin Kenfack
Ulrich Matchi Aïvodji
S Ebrahimi Kahou
Clinical prediction models are often personalized to target heterogeneous sub-groups by using demographic attributes such as race and gender… (see more) to train the model. Traditional personalization approaches involve using demographic attributes in input features or training multiple sub-models for different population subgroups (decoupling model). However, these methods often harm the performance at the subgroup level compared to non-personalized models. This paper presents a novel personalization method to improve model performance at the sub-group level. Our method involves a two-step process: first, we train a model to predict group attributes, and then we use this model to learn data-dependent biases to modulate a second model for diagnosis prediction. Our results demonstrate that this joint architecture achieves consistent performance gains across all sub-groups in the Heart dataset. Furthermore, in the mortality dataset, it improves performance in two of the four sub-groups. A comparison of our method with the traditional decoupled personalization method demonstrated a greater performance gain in the sub-groups with less harm. This approach offers a more effective and scalable solution for personalized models, which could have a positive impact in healthcare and other areas that require predictive models that take sub-group information into account.
2025-03-05
ICLR.cc/2025/Workshop/SCSL (published)
openreview.net
openreview.net
Towards Protein Sequence & Structure Co-Design with Multi-Modal Language Models
Stephen Zhewen Lu
Jiarui Lu
Hongyu Guo
Jian Tang
Proteins perform diverse biological functions, governed by the intricate relationship between their sequence and three-dimensional structure… (see more). While protein language models (PLMs) have demonstrated remarkable success in functional annotation and structure prediction, their potential for sequence-structure co-design remains underexplored. This limitation arises from pre-training objectives that favor masked token prediction over generative modeling. In this work, we systematically explore sampling strategies to enhance the generative capabilities of PLMs for co-design. Notably, we introduce a ranked iterative decoding with re-masking scheme, enabling PLMs to generate sequences and structures more effectively. Benchmarking ESM3 across multiple scales, we demonstrate that using PLMs effectively at sampling time for co-design tasks can outperform specialized architectures that lack comparable scaling properties. Our work advances the field of computational protein design by equipping PLMs with robust generative capabilities tailored to sequence-structure interdependence.
2025-03-05
ICLR.cc/2025/Workshop/LMRL (published)
openreview.net
openreview.net
Understanding (Un)Reliability of Steering Vectors in Language Models
Joschka Braun
Carsten Eickhoff
David M. Krueger
Seyed Ali Bahrainian
Dmitrii Krasheninnikov
Steering vectors are a lightweight method to control language model behavior by adding a learned bias to the activations at inference time. … (see more)Although steering demonstrates promising performance, recent work shows that it can be unreliable or even counterproductive in some cases. This paper studies the influence of prompt types and the geometry of activation differences on steering reliability. First, we find that all seven prompt types used in our experiments produce a net positive steering effect, but exhibit high variance across samples, and often give an effect opposite of the desired one. No prompt type clearly outperforms the others, and yet the steering vectors resulting from the different prompt types often differ directionally (as measured by cosine similarity). Second, we show that higher cosine similarity between training set activation differences predicts more effective steering. Finally, we observe that datasets where positive and negative activations are better separated are more steerable. Our results suggest that vector steering is unreliable when the target behavior is not represented by a coherent direction.
2025-03-05
ICLR.cc/2025/Workshop/FM-Wild (published)
doi.org
openreview.net
Who is your ideal peer mentor? A qualitative study to identify cancer patient preferences for a digital peer support app
Loes Knaapen
Andrea M. Laizner
Kelly Agnew
Xiao Jian Du
Douaa El Abiad
Luc Galarneau
Susie Judd
James Manalad
Ridhi Mittal
Tristan Williams
Brandon Woolfson
Angele Wen
J. Kildea
2025-03-05
Supportive Care in Cancer (published)
doi.org
Cracking the Code of Action: A Generative Approach to Affordances for Reinforcement Learning
Lynn Cherif
Flemming Kondrup
David Venuto
Ankit Anand
Doina Precup
Khimya Khetarpal
Agents that can autonomously navigate the web through a graphical user interface (GUI) using a unified action space (e.g., mouse and keyboar… (see more)d actions) can require very large amounts of domain-specific expert demonstrations to achieve good performance. Low sample efficiency is often exacerbated in sparse-reward and large-action-space environments, such as a web GUI, where only a few actions are relevant in any given situation. In this work, we consider the low-data regime, with limited or no access to expert behavior. To enable sample-efficient learning, we explore the effect of constraining the action space through
2025-03-04
ICLR.cc/2025/Workshop/DL4C (published)
doi.org
openreview.net
Design of Ligand-Binding Proteins with Atomic Flow Matching
Junqi Liu
Shaoning Li
Chence Shi
Zhi Yang
Jian Tang
2025-03-04
ICLR.cc/2025/Workshop/GEM (published)
openreview.net
openreview.net
Foundation models for generalizable electrocardiogram interpretation: comparison of supervised and self-supervised electrocardiogram foundation models
Alexis Nolin-Lapalme
Achille Sowa
Jacques Delfrate
Olivier Tastet
Denis Corbin
Merve Kulbay
Derman Ozdemir
Marie-Jeanne Noël
François-Christophe Marois-Blanchet
François Harvey
Surbhi Sharma
Minhaj Ansari
I-Min Chiu
Valentina Dsouza
Sam F. Friedman
Michael Chassé
Brian J. Potter
Jonathan Afilalo
Pierre Adil Elias
Gilbert Jabbour … (see 13 more)
Mourad Bahani
Marie-Pierre Dubé
Patrick M. Boyle
Neal A. Chatterjee
Joshua Barrios
Geoffrey H. Tison
David Ouyang
Mahnaz Maddah
Shaan Khurshid
Julia Cadrin-Tourigny
Rafik Tadros
Julie Hussin
Robert Avram
The 12-lead electrocardiogram (ECG) remains a cornerstone of cardiac diagnostics, yet existing artificial intelligence (AI) solutions for au… (see more)tomated interpretation often lack generalizability, remain closed-source, and are primarily trained using supervised learning, limiting their adaptability across diverse clinical settings. To address these challenges, we developed and compared two open-source foundational ECG models: DeepECG-SSL, a self-supervised learning model, and DeepECG-SL, a supervised learning model. Both models were trained on over 1 million ECGs using a standardized preprocessing pipeline and automated free-text extraction from ECG reports to predict 77 cardiac conditions. DeepECG-SSL was pretrained using self-supervised contrastive learning and masked lead modeling. The models were evaluated on six multilingual private healthcare systems and four public datasets for ECG interpretation across 77 diagnostic categories. Fairness analyses assessed disparities in performance across age and sex groups, while also investigating fairness and resource utilization. DeepECG-SSL achieved AUROCs of 0.990 (95%CI 0.990, 0.990) on internal dataset, 0.981 (95%CI 0.981, 0.981) on external public datasets, and 0.983 (95%CI 0.983, 0.983) on external private datasets, while DeepECG-SL demonstrated AUROCs of 0.992 (95%CI 0.992, 0.992), 0.980 (95%CI 0.980, 0.980) and 0.983 (95%CI 0.983, 0.983) respectively. Fairness analyses revealed minimal disparities (true positive rate & false positive rate difference<0.010) across age and sex groups. Digital biomarker prediction (Long QT syndrome (LQTS) classification, 5-year atrial fibrillation prediction and left ventricular ejection fraction (LVEF) classification) with limited labeled data, DeepECG-SSL outperformed DeepECG-SL in predicting 5-year atrial fibrillation risk (N=132,050; AUROC 0.742 vs. 0.720; Δ=0.022; P<0.001), identifying reduced LVEF ≤40% (N=25,252; 0.928 vs. 0.900; Δ=0.028; P<0.001), and classifying LQTS syndrome subtypes (N=127; 0.931 vs. 0.853; Δ=0.078; P=0.026). By releasing model weights, preprocessing tools, and validation code, we aim to support robust, data-efficient AI diagnostics across diverse clinical environments. This study establishes self-supervised learning as a promising paradigm for ECG analysis, particularly in settings with limited annotated data, enhancing accessibility, generalizability, and fairness in AI-driven cardiac diagnostics. Can self-supervised (SSL) learning yield ECG-based AI foundational models with enhanced performance, fairness, privacy, and generalizability compared to traditional supervised learning (SL) approaches? Our evaluation of DeepECG-SL and DeepECG-SSL across seven external health center datasets and four international publicly accessible datasets demonstrated that while both models achieve comparable diagnostic accuracy for ECG interpretation, SSL outperforms SL on novel tasks with smaller datasets. We validated DeepECG-SL and DeepECG-SSL across public and private datasets and demonstrated that SSL model had a superior generalizability by addressing fairness, privacy, and efficiency, and open sourcing our models, we advance ethical, adaptable AI for equitable, real-world ECG diagnostics. Graphical abstract: DeepECG-SL and DeepECG-SSL, two open-source AI models for 12-lead ECG interpretation, were trained on over 1 million ECGs. DeepECG-SSL, utilizing self-supervised contrastive learning and masked lead modeling, outperformed DeepECG-SL in utilizing digital biomarkers to predict atrial fibrillation risk, reduced LVEF, and long QT syndrome subtypes, while both models achieved high diagnostic accuracy with minimal fairness disparities across age and sex. Validated on ten external datasets, our work provides a robust, reproducible framework for equitable, efficient ECG-based cardiac diagnostics.
2025-03-04
medRxiv (preprint)
doi.org