On Generalization in Bilevel Optimization with Overparameterized Models

Fares El Khoury, Edouard Pauwels, Samuel Vaiter, Michael Arbel

Sydney Poster Session 2, Tue, Dec 8, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

Learning Global Probabilistic Explanations

Frederic Koriche, Louenas Bounia

Paris Poster Session 5, Fri, Dec 11, 11:30 AM–1:30 PM, Paris Poster Hall · Published 2026

BlockFormer: Transformer-based inference from interaction maps

Eloïse Touron, Pedro Rodrigues, Julyan Arbel, Nelle Varoquaux and 1 more

Paris Poster Session 5, Fri, Dec 11, 11:30 AM–1:30 PM, Paris Poster Hall · Published 2026

Validating Causal Abstraction Metrics on Simulated Complex Systems

Benchmarking thirty metrics on ten simulated complex systems shows only causal metrics reliably validate high-level explanations when testing unmapped-variable faithfulness, leading to the Causal Abstraction Error metric converging with thirty interventions.

Maxime Méloux, Tiago Pimentel, François Portet, Maxime Peyrard

Paris Poster Session 5, Fri, Dec 11, 11:30 AM–1:30 PM, Paris Poster Hall · Published 2026

Sequential Membership Inference Attacks

Sequential membership inference attacks exploit model update sequences and canary insertion timing to achieve tighter privacy audits with higher attack power than single-model baselines.

Thomas Michel, Debabrota Basu, Emilie Kaufmann

Sydney Poster Session 4, Wed, Dec 9, 5:00 PM–8:00 PM, Hall 1-4 · Published 2026

A Generalized Tikhonov Layer for Interpretable-by-design Graph Neural Networks

<|message_model|><|content_text|>The Tikhonov layer is an interpretable graph neural network layer whose learnable parameters directly reveal which node features and topological aspects drive predictions. Its closed-form propagation solves a generalized graph Tikhonov problem, yielding built-in expl

Nicolas Tremblay, Filippo Maria Bianchi, Benjamin Ricaud

Paris Poster Session 4, Thu, Dec 10, 5:30 PM–7:30 PM, Paris Poster Hall · Published 2026

Likelihood-free inference of phylogenetic tree posterior distributions

A likelihood-free neural network estimates phylogenetic tree posteriors via sequence pair encodings and subtree merges, outperforming likelihood-based methods especially for intractable evolutionary models.

Luc Blassel, Noémie Sauvage, Pierre Barrat-Charlaix, Bastien Boussau and 2 more

Paris Poster Session 4, Thu, Dec 10, 5:30 PM–7:30 PM, Paris Poster Hall · Published 2026

Boosting Brain-to-Image Decoding with TRIBE v2 Data Augmentation

TRIBE v2 synthetic fMRI augmentation improves brain-to-image decoding by up to 68%, though optimal synthetic-to-real ratios vary by dataset, and synthetic-only training achieves above-chance zero-shot decoding.

Yohann Benchetrit, Marlene Careil, Simon Dahan, Hubert Banville and 2 more

Paris Poster Session 1, Wed, Dec 9, 12:30 PM–2:30 PM, Paris Poster Hall · Published 2026

Learning to Sample From Diffusion Models via Inverse Reinforcement Learning

Inverse reinforcement learning trains diffusion sampling schedules by matching target behavior via policy gradients, cutting ImageNet-64 tuning costs up to 9x versus grid search with 16% inference overhead.

Constant Bourdrez, Alexandre Verine, Olivier Cappé

Sydney Poster Session 1, Tue, Dec 8, 10:00 AM–1:00 PM, Hall 1-4 · Published 2026

Maxitive Donsker-Varadhan Formulation for Possibilistic Variational Inference

A maxitive Donsker-Varadhan formulation enables possibilistic variational inference with practical CBOpt optimizers for competitive image classification.

Jasraj Singh, Shelvia Wongso, Jeremie Houssineau, Badr-Eddine Cherief-Abdellatif

Paris Poster Session 5, Fri, Dec 11, 11:30 AM–1:30 PM, Paris Poster Hall · Published 2026

PAC-Bayesian Bounds for Learning Partially Observed Stochastic Linear Time-Invariant State-Space Systems with Inputs and Sub-Gaussian Noise

PAC-Bayesian bounds relate expected and empirical prediction errors for partially observed LTI state-space systems with sub-Gaussian noise, yielding finite-sample guarantees for system identification and parameter estimation.

Mihaly Petreczky, Mohamad Al Ahdab, John Leth

Paris Poster Session 5, Fri, Dec 11, 11:30 AM–1:30 PM, Paris Poster Hall · Published 2026

i-DEQ: A stable inertial Deep Equilibrium model for image restoration

i-DEQ uses momentum in deep equilibrium fixed-point iterations for stable, accelerated image restoration with convergence guarantees and twofold faster inference.

Antonin Clerc, Marien Renaud, Baudouin Denis de Senneville, Nicolas Papadakis

Paris Poster Session 6, Fri, Dec 11, 2:30 PM–4:30 PM, Paris Poster Hall · Published 2026

Assessing Per-Sample Membership Inference Vulnerability without Retraining

Per-sample membership inference vulnerability is governed by a data-dependent geometric measure, yielding a surrogate score using only a single model that outperforms loss-based baselines at identifying high-risk training points.

Valentin Dorseuil, Jamal Atif, Olivier Cappé

Paris Poster Session 2, Wed, Dec 9, 5:00 PM–7:00 PM, Paris Poster Hall · Published 2026