Pablo Piantanida

clea.chataigner@mila.quebec

Biographie

Je suis professeur au CentraleSupélec de l'Université Paris-Saclay avec le Centre national français de la recherche scientifique (CNRS), et directeur du Laboratoire international sur les systèmes d'apprentissage (ILLS) qui regroupe l'Université McGill, l'École de technologie supérieure (ÉTS), Mila - Institut québécois d'intelligence artificielle, le Centre national français de la recherche scientifique (CNRS), l'Université Paris-Saclay et l'École CentraleSupélec.

Mes recherches portent sur l'application de techniques statistiques et de théorie de l'information avancées au domaine de l'apprentissage automatique. Je m'intéresse au développement de techniques rigoureuses basées sur des mesures et des concepts d'information pour construire des systèmes d'IA sûrs et fiables et établir la confiance dans leur comportement et leur robustesse, sécurisant ainsi leur utilisation dans la société. Mes principaux domaines d'expertise sont la théorie de l'information, la géométrie de l'information, la théorie de l'apprentissage, la protection de la vie privée, l'équité, avec des applications à la vision par ordinateur et au traitement du langage naturel.

J'ai fait mes études de premier cycle à l'université de Buenos Aires et j'ai poursuivi des études supérieures en mathématiques appliquées à l'université Paris-Saclay en France. Tout au long de ma carrière, j'ai également occupé des postes d'invité à l'INRIA, à l'Université de Montréal et à l'École de technologie supérieure (ÉTS), entre autres.

Mes recherches antérieures ont porté sur les domaines de la théorie de l'information au-delà de la compression distribuée, de la décision statistique, du codage universel des sources, de la coopération, de la rétroaction, du codage d'index, de la génération de clés, de la sécurité et de la protection des données.

Je donne des cours sur l'apprentissage automatique, la théorie de l'information et l'apprentissage profond, couvrant des sujets tels que la théorie de l'apprentissage statistique, les mesures de l'information, les principes statistiques des réseaux neuronaux.

Étudiants actuels

Chataigner Cléa

Doctorat - McGill

Superviseur⋅e principal⋅e :

Golnoosh Farnadi

Maxime Darrin

Doctorat - McGill

Superviseur⋅e principal⋅e :

Jackie Cheung

maxime.darrin@mila.quebec

Doctorat - École de technologie suprérieure

zahra.dehghani-tafti@mila.quebec

philippe.formont@mila.quebec

Philippe Formont

Doctorat - École de technologie suprérieure

fanny.jourdan@mila.quebec

Fanny JOURDAN

Postdoctorat - École de technologie suprérieure

Co-superviseur⋅e :

Jackie Cheung

Site web

guillermo.martin@mila.quebec

Guillermo Martin de Paula

Doctorat - Université Paris-Saclay

Chiara Roverato

Doctorat - McGill

Superviseur⋅e principal⋅e :

Mark Coates

chiara.roverato@mila.quebec

Matteo Sammut

Doctorat - Université Paris Dauphine-PSL

matteo.sammut@dauphine.eu

Amin Sghaier

Maîtrise recherche - École de technologie suprérieure

Co-superviseur⋅e :

Marco Pedersoli

amin.sghaier@mila.quebec

Nicolas Thome

Collaborateur·rice de recherche - Sorbonne Université

thome.nicolas@gmail.com

Site web

Publications

ECUAS: A family of metrics for principled evaluation of uncertainty-augmented systems

Lautaro Estienne

Erik Ernst

Matias Vera

LUCIANA FERRER

In high-stakes automated decision-making, access to predictive uncertainty is essential for enabling users -- human or downstream systems --… (voir plus) to accept or reject predictions based on application-specific cost trade-offs. Such uncertainty-augmented (UA) systems -- i.e., systems that output both predictions and uncertainty scores -- are currently being assessed in the literature in a variety of ways, using separate metrics to evaluate the predictions and the uncertainty scores, setting a cost function with a fixed rejection cost or integrating over a coverage-risk curve. We argue that these evaluation approaches are inadequate for assessing overall performance of the UA system for decision making under uncertainty and propose a novel family of metrics, ECUAS

2026-12-30

OpenReview (publié)

QueStER: Query Specification for Generative Keyword-Based Retrieval

Arthur SATOUF

Yuxuan Zong

Habiboulaye Amadou Boubacar

Benjamin Piwowarski

Generative retrieval (GR) differs from the traditional index–then–retrieve pipeline by storing relevance in model parameters and generat… (voir plus)ing retrieval cues directly from the query, but it can be brittle out of domain and expensive to scale. We introduce QueStER (QUEry SpecificaTion for gEnerative Keyword-Based Retrieval), which bridges GR and query reformulation by learning to generate explicit keyword-based search specifications. Given a user query, a lightweight LLM produces a keyword query that is executed by a standard retriever (BM25), combining the generalization benefits of generative query rewriting with the efficiency and scalability of lexical indexing. We train the rewriting policy with reinforcement learning techniques. Across in- and out-of-domain evaluations, QueStER consistently improves over BM25 and is competitive with neural IR baselines, while maintaining strong efficiency.

2026-12-30

OpenReview (publié)

Source-Free Class Relearning: Diagnosing Forgetting in Class Unlearning

Zahra Dehghani

Mohammadhadi Shateri

Class unlearning aims to remove a model's ability to recognize designated forget classes while preserving performance on retain classes. How… (voir plus)ever, low forget accuracy after unlearning does not necessarily mean the class structure has been erased. Approximate unlearning methods can alter classifier decision boundaries while leaving recoverable structure in the representation. Prior work has shown that forget classes can be recovered, but existing approaches require real forget or retain samples, auxiliary data, or reference checkpoints. We study class relearning in a strictly source-free setting, asking whether a forget class can be recovered through a classifier-head update using only the unlearned model. Our approach rests on a theoretical analysis establishing a sufficient alignment condition under which a single gradient step on a synthetic probe set increases the expected logit margin of the forget class. Building on this, we propose a white-box Source-Free Relearning Audit (SFRA), which generates candidate embeddings in representation space and uses model-guided confidence filtering to construct high-confidence retain probes and low-confidence boundary-adjacent probes that are relabelled as the forget class. Gaussian sampling and Softmax confidence are used by default, while ablations with alternative proposal distributions and uncertainty criteria show that recoverability is not specific to these choices. To quantify recoverability, we introduce the Relearning Score (RS), which jointly measures forget-class recovery and retain-accuracy preservation, and report class-matched

2026-09-01

arXiv (prépublication)

arxiv.org

A Universal Source-Free Class Unlearning Framework via Synthetic Embeddings

Zahra Dehghani Tafti

Mohammadhadi Shateri

Class unlearning in neural classifiers refers to selectively removing the model’s ability to recognize a target (forget) class by reshapin… (voir plus)g the decision boundaries. This is essential when taxonomies change, labels are corrected, or legal or ethical requirements mandate class removal. The objective is to preserve performance on the remaining (retain) classes while avoiding costly full retraining. Existing methods generally require access to the source, i.e., forget/retain data or a relevant surrogate dataset. This dependency limits their applicability in scenarios where access to source data is restricted or unavailable. Even the recent source-free class unlearning methods rely on generating samples in the data space, which is computationally expensive and not even essential for doing class unlearning. In this work, we propose a novel source-free class unlearning framework that enables existing unlearning methods to operate using only the deployed model. We show that, under assumptions on the forget loss with respect to logits, class unlearning can be performed source-free for any given neural classifier by utilizing randomly generated samples within the classifier’s intermediate space. Specifically, randomly generated embeddings pseudo-labeled by the model as belonging to the forget or retain classes can support effective source-free unlearning. Our analysis further shows that, under conditions on the forget loss and synthetic forget embeddings, minimizing the forget loss induces expected logit shifts consistent with class unlearning, without requiring a specific parametric form of the embedding distribution. We validate our framework on four backbone architectures, ResNet-18, ResNet-50, ViT-B/16, and Swin-T, across three benchmark datasets, CIFAR-10, CIFAR-100, and TinyImageNet. Our experimental results show that existing class unlearning methods can operate within our source-free framework, with minimal impact on their forgetting efficacy and retain class accuracy. The code is available at https://github.com/Yasaman-dt/Source_Free_Class_Unlearning.

2026-05-18

Transactions on Machine Learning Research (accepté)

MolRGen: A Training and Evaluation Setting for De Novo Molecular Generation with Reasonning Models

Philippe Formont

Maxime Darrin

Ismail Ben Ayed

Recent advances in reasoning-based large language models (LLMs) have demonstrated substantial improvements in complex problem-solving tasks.… (voir plus) Motivated by these advances, several works have explored the application of reasoning LLMs to drug discovery and molecular design. However, most existing approaches either focus on evaluation or rely on training setups that require ground-truth labels, such as molecule pairs with known property modifications. Such supervision is unavailable in \textit{de novo} molecular generation, where the objective is to generate novel molecules that optimize a desirability score without prior knowledge of high-scoring candidates. To bridge this gap, we introduce MolRGen, a large-scale benchmark and dataset for training and evaluating reasoning-based LLMs on \textit{de novo} molecular generation. Our contributions are threefold. First, we propose a setting to evaluate and train models for \textit{de novo} molecular generation and property prediction. Second, we introduce a novel diversity-aware top-

2026-03-17

arXiv (prépublication)

arxiv.org

Position: Auditing Is Not Evaluating; LLM Audit Requires Dynamic, Contextual, Budget-Aware and Reliable Evidence

Cléa Chataigner

Golnoosh Farnadi

Auditing large language models (LLMs) is increasingly urgent as these systems are deployed in high-stakes settings, yet existing evaluation … (voir plus)practices are ill-suited to meet auditing requirements. Directly repurposing standard evaluation tools can yield incomplete or misleading conclusions, e.g. overstating robustness when evidence comes from static prompts rather than adaptive, real-world interactions. This position paper argues that effective LLM audits must instead generate dynamic, context-sensitive, budget-aware, and reliable evidence. To support this position, we analyze how each of these principles can be operationalized through a four-component framework: Auditing Scope, Interactor, Evaluator, and Output. We highlight design requirements, assumptions, limitations and research directions, demonstrating how high-level principles can be translated into concrete, actionable, evidence-based procedures.

2026-02-05

2026 (prépublication)

openalex.org

Happiness as a Measure of Fairness

Georg Pichler

Marco Romanelli

In this paper, we propose a novel fairness framework grounded in the concept of _happiness_, a measure of the utility each group gains from … (voir plus)decision outcomes. By capturing fairness through this intuitive lens, we not only offer a more human-centered approach, but also one that is mathematically rigorous: In order to compute the optimal, fair post-processing strategy, only a linear program needs to be solved. This makes our method both efficient and scalable with existing optimization tools. Furthermore, it unifies and extends several well-known fairness definitions, and our empirical results highlight its practical strengths across diverse scenarios.

2026-02-02

Artificial Intelligence and Statistics (poster)

An Indicator of Membership Inference Security in Post-Training Quantized Models

Eric Aubinais

Philippe Formont

Elisabeth Gassiat

Quantizing machine learning models has demonstrated its effectiveness in lowering memory and inference costs while maintaining performance l… (voir plus)evels comparable to those of the original models. In this work, we investigate the impact of quantization procedures on privacy in data-driven models, focusing on their vulnerability to membership inference attacks. Membership Inference Security (MIS) has recently been proposed to characterize the privacy of machine learning models against the most powerful (and possibly unknown) attacks. However, quantifying MIS appears to be computationally very difficult. In this paper, we propose a new MIS indicator for post-training quantization procedures of machine learning models that minimize an empirical loss. This new indicator is a byproduct of a theoretical asymptotic analysis of the MIS in this context. We also present a methodology for empirically estimating our MIS indicator. Using synthetic datasets and real-world data (in the context of drug discovery), we demonstrate the effectiveness of our approach in assessing and ranking the MIS of different quantizers.

2026-02-02

International Conference on Artificial Intelligence and Statistics (poster)

Adapting Language Models to Produce Good Class Probabilities for Classification Tasks

Lautaro Estienne

Matias Vera

Elizabeth Fons

Elena Kochkina

LUCIANA FERRER

Large generative language models (GLM) provide a versatile tool for solving a wide variety of natural processing tasks. GLM responses, thoug… (voir plus)h, are provided in the form of text, without an indication of the model's confidence in the answer. This limits the usability of these models on high-risk applications where decisions made based on an incorrect answer can have severe consequences. In this work, we focus on the problem of generating class posterior distributions for text classification tasks like sentiment, news category and intent classification. These posteriors can be used for decision making and as interpretable scores for the user. We show that the naive approach for computing posteriors based on the token posteriors produced by the GLM results in extremely poor posteriors. We then explore different adaptation approaches for improving the quality of posteriors, focusing on low resource scenarios where a small amount of data is available for adaptation. We show that parameter-efficient supervised fine-tuning (SFT), while providing large gains in terms of decision quality, produces suboptimal posteriors due to overfitting. To address this problem, we propose an approach that combines SFT and post-hoc calibration (PHC) using a three-stage training strategy, improving the quality of both posteriors and categorical decisions.

2026-01-31

Transactions on Machine Learning Research (accepté)

BayesAdapter: enhanced uncertainty estimation in CLIP few-shot adaptation

Pablo Morales-Álvarez

Stergios Christodoulidis

Maria Vakalopoulou

Jose Dolz

The emergence of large pre-trained vision-language models (VLMs) represents a paradigm shift in machine learning, with unprecedented results… (voir plus) in a broad span of visual recognition tasks. CLIP, one of the most popular VLMs, has exhibited remarkable zero-shot and transfer learning capabilities in classification. To transfer CLIP to downstream tasks, adapters constitute a parameter-efficient approach that avoids backpropagation through the large model (unlike related prompt learning methods). However, CLIP adapters have been developed to target discriminative performance, and the quality of their uncertainty estimates has been overlooked. In this work we show that the discriminative performance of state-of-the-art CLIP adapters does not always correlate with their uncertainty estimation capabilities, which are essential for a safe deployment in real-world scenarios. We also demonstrate that one of such adapters is obtained through MAP inference from a more general probabilistic framework. Based on this observation we introduce BayesAdapter, which leverages Bayesian inference to estimate a full probability distribution instead of a single point, better capturing the variability inherent in the parameter space. In a comprehensive empirical evaluation we show that our approach obtains high quality uncertainty estimates in the predictions, standing out in calibration and selective classification. Our code will be publicly available upon acceptance of the paper.

2026-01-09

International Journal of Computer Vision (publié)

arxiv.org

From Infrastructure to Interface, the AI Value Chain Drives LLM Homogenization

Cléa Chataigner

Jackie CK Cheung

2025-12-31

Neural Information Processing Systems (Accept)

Learning Task-Agnostic Representations through Multi-Teacher Distillation

Jackie CK Cheung

Ismail Ben Ayed

Mohammadhadi Shateri

Casting complex inputs into tractable representations is a critical step across various fields. Diverse embedding models emerge from differe… (voir plus)nces in architectures, loss functions, input modalities and datasets, each capturing unique aspects of the input. Multi-teacher distillation leverages this diversity to enrich representations but often remains tailored to specific tasks. In this paper, we introduce a task-agnostic framework based on a ``majority vote" objective function. We demonstrate that this function is bounded by the mutual information between student and teachers' embeddings, leading to a task-agnostic distillation loss that eliminates dependence on task-specific labels or prior knowledge. Our evaluations across text, vision models, and molecular modeling show that our method effectively leverages teacher diversity, resulting in representations enabling better performance for a wide range of downstream tasks such as classification, clustering, or regression. Additionally, we train and release state-of-the-art embedding models, enhancing downstream performance in various modalities.

2025-12-30

OpenReview (publié)