7.1.1 declearn

• Keyword:
  Federated learning
• Scientific Description:

  declearn is a python package providing with a framework to perform federated learning, i.e. to train machine learning models by distributing computations across a set of data owners that, consequently, only have to share aggregated information (rather than individual data samples) with an orchestrating server (and, by extension, with each other).

  The aim of declearn is to provide both real-world end-users and algorithm researchers with a modular and extensible framework that:

  (1) builds on abstractions general enough to write backbone algorithmic code agnostic to the actual computation framework, statistical model details or network communications setup

  (2) designs modular and combinable objects, so that algorithmic features, and more generally any specific implementation of a component (the model, network protocol, client or server optimizer...) may easily be plugged into the main federated learning process - enabling users to experiment with configurations that intersect unitary features

  (3) provides with functioning tools that may be used out-of-the-box to set up federated learning tasks using some popular computation frameworks (scikit- learn, tensorflow, pytorch...) and federated learning algorithms (FedAvg, Scaffold, FedYogi...)

  (4) provides with tools that enable extending the support of existing tools and APIs to custom functions and classes without having to hack into the source code, merely adding new features (tensor libraries, model classes, optimization plug-ins, orchestration algorithms, communication protocols...) to the party.

  Parts of the declearn code (Optimizers,...) are included in the FedBioMed software.

  At the moment, declearn has been focused on so-called "centralized" federated learning that implies a central server orchestrating computations, but it might become more oriented towards decentralized processes in the future, that remove the use of a central agent.

• Functional Description:

  This library provides the two main components to perform federated learning:

  (1) the client, to be run by each participant, performs the learning on local data et releases only the result of the computation

  (2) the server orchestrates the process and aggregates the local models in a global model

• News of the Year:
  Two major releases with key new functionalities including algorithms for group fairness and the ability to use secure aggregation.
• URL:
  https://­gitlab.­inria.­fr/­magnet/­declearn/­declearn2
• Contact:
  Aurélien Bellet
• Participants:
  Paul Andrey, Aurélien Bellet, Nathan Bigaud, Marc Tommasi, Nathalie Vauquier
• Partner:
  CHRU Lille

Results: Choice of the causal measure 2

Participants: Julie Josse.

There are many measures to report so-called treatment or causal effect: absolute difference, ratio, odds ratio, number needed to treat, and so on. The choice of a measure, e.g. absolute versus relative, is often debated because it leads to different appreciations of the same phenomenon; but it also implies different heterogeneity of treatment effect. In addition some measures – but not all – have appealing properties such as collapsibility, matching the intuition of a population summary. We review common measures and their pros and cons typically brought forward. Doing so, we clarify notions of collapsibility and treatment effect heterogeneity, unifying different existing definitions. Our main contribution is to propose to reverse the thinking: rather than starting from the measure, we start from a non-parametric generative model of the outcome. Depending on the nature of the outcome, some causal measures disentangle treatment modulations from baseline risk. Therefore, our analysis outlines an understanding of what heterogeneity and homogeneity of treatment effect mean, not through the lens of the measure, but through the lens of the covariates. Our goal is the generalization of causal measures. We show that different sets of covariates are needed to generalize an effect to a different target population depending on (i) the causal measure of interest, (ii) the nature of the outcome, and (iii) the generalization’s method itself (generalizing either conditional outcome or local effects).

Results: Federated Causal Inference 36

Participants: Remi Khellaf, Aurélien Bellet, Julie Josse.

Randomized Controlled Trials (RCTs) are the gold standard for estimating the Average Treatment Effect (ATE) in evidence-based medicine, but their limitations—such as stringent eligibility criteria and small sample sizes—have led to the prominence of meta-analyses, the pinnacle of evidence in clinical research, which aggregate evidence from multiple studies to enhance statistical power and precision.

Despite extensive guidelines on conducting meta-analyses, multi-centric approaches still face significant challenges. These primarily arise from heterogeneity caused by imbalances in datasets, variations in populations across studies, and center effects due to differing practices across institutions. Moreover, simply aggregating local estimates is not the only approach to conducting meta-analyses. However, implementing “one-stage” meta-analyses that pool individual patient data from all centers is practically challenging due to data silos and personal data regulations.

Federated causal inference offers a promising alternative by allowing decentralized data sources to collaborate without sharing raw data, thus maintaining privacy and compliance with regulations. This work investigates three federated ATE estimation approaches—meta-analysis estimators, one-shot federated estimators, and gradient-based federated estimators—comparing their trade-offs in statistical efficiency, communication costs, and robustness to heterogeneity. The study demonstrates that meta-analysis estimators can achieve statistical efficiency comparable to pooled data analysis when sufficient data is available at each center, while naturally accommodating center effects. In contrast, while gradient-based approaches excel in low-data scenarios, one-shot estimators can be robust to distributional shifts but suffer from increased variance when center effects are present.

Guidelines and a decision diagram are provided to help practitioners choose the most appropriate approach based on data and heterogeneity conditions.

Results: Distribution on Distribution Regression to model Treatment Response Assessment in Asthma Patients

Participants: Marie Felicia Beclin, Nicolas Molinari.

Medical imaging plays a crucial role in evaluating treatment efficacy. While practitioners traditionally rely on specific biomarkers and clinical data, incorporating informative features derived from medical imaging can enhance treatment response prediction. This research focuses on thoracic scans taken in expiration and inspiration before and after one year of Benralizumab treatment for asthma patients.

Following image segmentation, histograms are calculated to represent the distribution of voxel intensities. The underlying hypothesis posits that patients with improved conditions will exhibit enhanced expiration scans after treatment, evident in the histograms through a rightward shift, indicating higher Hounsfield Unit (HU) values. To predict treatment's response, we develop an histogram on histogram regression. Unlike existing methods, our proposed model goes beyond point-wise estimation of coefficient, offering an inferential framework to obtain p-values and confidence intervals for assessing treatment effects.

Results: Missing values imputation 41

Participants: Julie Josse, Jeffrey Naef.

Missing values pose a persistent challenge in modern data science. Consequently, there is an ever-growing number of publications introducing new imputation methods in various fields. The present paper attempts to take a step back and provide a more systematic analysis. Starting from an in-depth discussion of the Missing at Random (MAR) condition for nonparametric imputation, we first develop an identification result, showing that the widely used Multiple Imputation by Chained Equations (MICE) approach indeed identifies the right conditional distributions. Building on this analysis, we propose three essential properties a successful imputation method should meet, thus enabling a more principled evaluation of existing methods and more targeted development of new methods. In particular, we introduce a new imputation method, denoted mice-DRF, that meets two out of the three criteria. We then discuss and refine ways to rank imputation methods, developing a powerful, easy-to-use scoring algorithm to rank missing value imputations.

Results: Conformal prediction with missing values 47

Participants: Margaux Zaffran, Julie Josse.

By leveraging increasingly large data sets, statistical algorithms and machine learning methods can be used to support, high-stakes decision-making problems such as autonomous driving, medical or civic applications, and more. To ensure the safe deployment of predictive models, it is crucial to quantify the uncertainty of the resulting predictions, communicating the limits of predictive performance. Uncertainty quantification attracts a lot of attention in recent years, particularly methods that are based on Conformal Prediction.

We investigate how to adequately quantify predictive uncertainty with missing covariates. A bottleneck is that missing values induce heteroskedasticity on the response's predictive distribution given the observed covariates. Thus, we focus on building predictive sets for the response that are valid conditionally to the missing values pattern. We show that this goal is impossible to achieve informatively in a distribution-free fashion, and we propose useful restrictions on the distribution class. Motivated by these hardness results, we characterize how missing values and predictive uncertainty intertwine. Particularly, we rigorously formalize the idea that the more missing values, the higher the predictive uncertainty. Then, we introduce a generalized framework, coined CP-MDA-Nested, outputting predictive sets in both regression and classification. Under independence between the missing value pattern and both the features and the response (an assumption justified by our hardness results), these predictive sets are valid conditionally to any pattern of missing values. Moreover, it provides great flexibility in the trade-off between statistical variability and efficiency. Finally, we experimentally assess the performances of CP-MDA-Nested beyond its scope of theoretical validity, demonstrating promising outcomes in more challenging configurations than independence.

Results: Rényi Pufferfish Privacy 27

Participants: Aurélien Bellet.

Pufferfish privacy is a flexible generalization of differential privacy that allows to model arbitrary secrets and adversary's prior knowledge about the data (e.g., correlation across individuals). Unfortunately, designing general and tractable Pufferfish mechanisms that do not compromise utility is challenging. Furthermore, this framework does not provide the composition guarantees needed for a direct use in iterative machine learning algorithms. To mitigate these issues, we introduce a Rényi divergence-based variant of Pufferfish and show that it allows us to extend the applicability of the Pufferfish framework. We first generalize the Wasserstein mechanism to cover a wide range of noise distributions and introduce several ways to improve its utility. We also derive stronger guarantees against out-of-distribution adversaries. Finally, as an alternative to composition, we prove privacy amplification results for contractive noisy iterations and showcase the first use of Pufferfish in private convex optimization. A common ingredient underlying our results is the use and extension of shift reduction lemmas.

Results: Relative Gaussian Mechanism 24

Participants: Aurélien Bellet.

The Gaussian Mechanism (GM), which consists in adding Gaussian noise to a vector-valued query before releasing it, is a standard privacy protection mechanism. In particular, given that the query respects some L2 sensitivity property (the L2 distance between outputs on any two neighboring inputs is bounded), GM guarantees Rényi Differential Privacy (RDP). Unfortunately, precisely bounding the L2 sensitivity can be hard, thus leading to loose privacy bounds. In this work, we consider a Relative L2 sensitivity assumption, in which the bound on the distance between two query outputs may also depend on their norm. Leveraging this assumption, we introduce the Relative Gaussian Mechanism (RGM), in which the variance of the noise depends on the norm of the output. We prove tight bounds on the RDP parameters under relative L2 sensitivity, and characterize the privacy loss incurred by using output-dependent noise. In particular, we show that RGM naturally adapts to a latent variable that would control the norm of the output. Finally, we instantiate our framework to show tight guarantees for Private Gradient Descent, a problem that naturally fits our relative L2 sensitivity assumption.

Results: Confidential Proof of Differentially Private Training 29

Participants: Ioan Tudor Cebere, Aurélien Bellet.

Post hoc privacy auditing techniques can be used to test the privacy guarantees of a model, but come with several limitations: (i) they can only establish lower bounds on the privacy loss, (ii) the intermediate model updates and some data must be shared with the auditor to get a better approximation of the privacy loss, and (iii) the auditor typically faces a steep computational cost to run a large number of attacks. In this paper, we propose to proactively generate a cryptographic certificate of privacy during training to forego such auditing limitations. We introduce Confidential-DPproof, a framework for Confidential Proof of Differentially Private Training, which enhances training with a certificate of the ($ϵ$, $\delta$)-DP guarantee achieved. To obtain this certificate without revealing information about the training data or model, we design a customized zero-knowledge proof protocol tailored to the requirements introduced by differentially private training, including random noise addition and privacy amplification by subsampling. In experiments on CIFAR-10, Confidential-DPproof trains a model achieving state-of-the-art 91% test accuracy with a certified privacy guarantee of ($ϵ=0.55$,$\delta =10-5$)-DP in approximately 100 hours.

Results: Private Training of Lipschitz Neural Networks 21

Participants: Aurélien Bellet.

State-of-the-art approaches for training Differentially Private (DP) Deep Neural Networks (DNN) face difficulties to estimate tight bounds on the sensitivity of the network's layers, and instead rely on a process of per-sample gradient clipping. This clipping process not only biases the direction of gradients but also proves costly both in memory consumption and in computation. To provide sensitivity bounds and bypass the drawbacks of the clipping process, we propose to rely on Lipschitz constrained networks. Our theoretical analysis reveals an unexplored link between the Lipschitz constant with respect to their input and the one with respect to their parameters. By bounding the Lipschitz constant of each layer with respect to its parameters, we prove that we can train these networks with privacy guarantees. Our analysis not only allows the computation of the aforementioned sensitivities at scale, but also provides guidance on how to maximize the gradient-to-noise ratio for fixed privacy guarantees. The code has been released as a Python package.

Results: Private Decentralized Learning with Random Walks 22

Participants: Aurélien Bellet.

The popularity of federated learning comes from the possibility of better scalability and the ability for participants to keep control of their data, improving data security and sovereignty. Unfortunately, sharing model updates also creates a new privacy attack surface. In this work, we characterize the privacy guarantees of decentralized learning with random walk algorithms, where a model is updated by traveling from one node to another along the edges of a communication graph. Using a recent variant of differential privacy tailored to the study of decentralized algorithms, namely Pairwise Network Differential Privacy, we derive closed-form expressions for the privacy loss between each pair of nodes where the impact of the communication topology is captured by graph theoretic quantities. Our results further reveal that random walk algorithms tends to yield better privacy guarantees than gossip algorithms for nodes close from each other. We supplement our theoretical results with empirical evaluation on synthetic and real-world graphs and datasets.

Results: Privacy Attacks in Decentralized Learning 26

Participants: Aurélien Bellet.

Decentralized Gradient Descent (D-GD) allows a set of users to perform collaborative learning without sharing their data by iteratively averaging local model updates with their neighbors in a network graph. The absence of direct communication between non-neighbor nodes might lead to the belief that users cannot infer precise information about the data of others. In this work, we demonstrate the opposite, by proposing the first attack against D-GD that enables a user (or set of users) to reconstruct the private data of other users outside their immediate neighborhood. Our approach is based on a reconstruction attack against the gossip averaging protocol, which we then extend to handle the additional challenges raised by D-GD. We validate the effectiveness of our attack on real graphs and datasets, showing that the number of users compromised by a single or a handful of attackers is often surprisingly large. We empirically investigate some of the factors that affect the performance of the attack, namely the graph topology, the number of attackers, and their position in the graph.

Results: Privacy Auditing of Machine Learning 33

Participants: Ioan Tudor Cebere, Aurélien Bellet.

Machine learning models can be trained with formal privacy guarantees via differentially private optimizers such as Differential Privacy Stochastic Gradient Descent (DP-SGD). In this work, we focus on a threat model where the adversary has access only to the final model, with no visibility into intermediate updates. In the literature, this "hidden state" threat model exhibits a significant gap between the lower bound from empirical privacy auditing and the theoretical upper bound provided by privacy accounting. To challenge this gap, we propose to audit this threat model with adversaries that craft a gradient sequence designed to maximize the privacy loss of the final model without relying on intermediate updates. Our experiments show that this approach consistently outperforms previous attempts at auditing the hidden state model. Furthermore, our results advance the understanding of achievable privacy guarantees within this threat model. Specifically, when the crafted gradient is inserted at every optimization step, we show that concealing the intermediate model updates in DP-SGD does not amplify privacy. The situation is more complex when the crafted gradient is not inserted at every step: our auditing lower bound matches the privacy upper bound only for an adversarially-chosen loss landscape and a sufficiently large batch size. This suggests that existing privacy upper bounds can be improved in certain regimes.

Results: Private Histogram Estimation 43

Participants: Aurélien Bellet.

We present Nebula, a system for differential private histogram estimation of data distributed among clients. Nebula enables clients to locally subsample and encode their data such that an untrusted server learns only data values that meet an aggregation threshold to satisfy differential privacy guarantees. Compared with other private histogram estimation systems, Nebula uniquely achieves all of the following: i) a strict upper bound on privacy leakage; ii) client privacy under realistic trust assumptions; iii) significantly better utility compared to standard local differential privacy systems; and iv) avoiding trusted third-parties, multi-party computation, or trusted hardware. We provide both a formal evaluation of Nebula's privacy, utility and efficiency guarantees, along with an empirical evaluation on three real-world datasets. We demonstrate that clients can encode and upload their data efficiently (only 0.0058 seconds running time and 0.0027 MB data communication) and privately (strong differential privacy guarantees ε = 1). On the United States Census dataset, the Nebula's untrusted aggregation server estimates histograms with above 88% better utility than the existing local deployment of differential privacy. Additionally, we describe a variant that allows clients to submit multi-dimensional data, with similar privacy, utility, and performance. Finally, we provide an open source implementation of Nebula.

Results: Correlated Gaussian Mechanism 37

Participants: Christian Janos Lebeda.

We consider the problem of releasing a sparse histogram under (ε,δ)-differential privacy. The stability histogram independently adds noise from a Laplace or Gaussian distribution to the non-zero entries and removes those noisy counts below a threshold. Thereby, the introduction of new non-zero values between neighboring histograms is only revealed with probability at most δ, and typically, the value of the threshold dominates the error of the mechanism. We consider the variant of the stability histogram with Gaussian noise. Recent works reduced the error for private histograms using correlated Gaussian noise. However, these techniques can not be directly applied in the very sparse setting. Instead, we adopt Lebeda's technique and show that adding correlated noise to the non-zero counts only allows us to reduce the magnitude of noise when we have a sparsity bound. This, in turn, allows us to use a lower threshold by up to a factor of 1/2 compared to the non-correlated noise mechanism. We then extend our mechanism to a setting without a known bound on sparsity. Additionally, we show that correlated noise can give a similar improvement for the more practical discrete Gaussian mechanism.

Results: Generalization Guarantees for Decentralized SGD 25

Participants: Aurélien Bellet.

This work presents a new generalization error analysis for Decentralized Stochastic Gradient Descent (D-SGD) based on algorithmic stability. The obtained results overhaul a series of recent works that suggested an increased instability due to decentralization and a detrimental impact of poorly-connected communication graphs on generalization. On the contrary, we show, for convex, strongly convex and non-convex functions, that D-SGD can always recover generalization bounds analogous to those of classical SGD, suggesting that the choice of graph does not matter. We then argue that this result is coming from a worst-case analysis, and we provide a refined optimization-dependent generalization bound for general convex functions. This new bound reveals that the choice of graph can in fact improve the worst-case bound in certain regimes, and that surprisingly, a poorly-connected graph can even be beneficial for generalization.

Results: Federated Conformal Prediction 35

Participants: Aurélien Bellet.

We study conformal prediction in the one-shot federated learning setting. The main goal is to compute marginally and training-conditionally valid prediction sets, at the server-level, in only one round of communication between the agents and the server. Using the quantile-of-quantiles family of estimators and split conformal prediction, we introduce a collection of computationally-efficient and distribution-free algorithms that satisfy the aforementioned requirements. Our approaches come from theoretical results related to order statistics and the analysis of the Beta-Beta distribution. We also prove upper bounds on the coverage of all proposed algorithms when the nonconformity scores are almost surely distinct. For algorithms with training-conditional guarantees, these bounds are of the same order of magnitude as those of the centralized case. Remarkably, this implies that the one-shot federated learning setting entails no significant loss compared to the centralized case. Our experiments confirm that our algorithms return prediction sets with coverage and length similar to those obtained in a centralized setting.

Results: Synthetic Data Generation for Intersectional Fairness 39

Participants: Aurélien Bellet.

In this work, we introduce a data augmentation approach specifically tailored to enhance intersectional fairness in classification tasks. Our method capitalizes on the hierarchical structure inherent to intersectionality, by viewing groups as intersections of their parent categories. This perspective allows us to augment data for smaller groups by learning a transformation function that combines data from these parent groups. Our empirical analysis, conducted on four diverse datasets including both text and images, reveals that classifiers trained with this data augmentation approach achieve superior intersectional fairness and are more robust to "leveling down" when compared to methods optimizing traditional group fairness metrics.

Results: Probabilistic Prediction of Arrivals and Hospitalizations in Emergency Departments in Île-de-France 45

Adaptive probabilistic forecasting of French electricity spot prices

Background: Forecasts of future demand is foundational for effective resource allocation in emergency departments (EDs). As ED demand is inherently variable, it is important for forecasts to characterize the range of possible future demand. However, extant research focuses primarily on producing point forecasts using a wide variety of prediction algorithms. In this study, our objective is to generate point and interval predictions that accurately characterize the variability in ED demand using ensemble methods that combine predictions from multiple base algorithms based on their empirical performance.

Methods: Data consisted in daily arrivals and subsequent hospitalizations at 72 emergency departments in Ile-de-France from 2014-2018. Additional explanatory variables were collected including public and school holidays, meteorological variables, and public health trends. One-day ahead point and 80% interval pre- dictions of arrivals and hospitalizations were produced by predicting the 10%, 50%, and 90% quantiles of the forecast distribution. Quantile prediction algorithms included methods such as ARIMAX, variations of random forests, and generalized additive models. Ensemble predictions were then formed using Exponentially Weighted Averaging, Bernstein Online Aggregation, and Super Learning. Prediction intervals were post-processed using Adaptive Conformal Inference techniques. Point predictions were evaluated by their Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE), and 80% interval predictions by their empirical coverage and mean interval width.

Results: For point forecasts, ensemble methods achieved lower average MAE and MAPE than any of the base algorithms. All of the base algorithms and ensemble methods yielded prediction intervals with near optimal empirical coverage after conformalization. For hospitalizations, the shortest mean interval widths were achieved by the ensemble methods.

Conclusions: Ensemble methods yield joint point and prediction intervals that adapt to individual EDs and achieve better performance than individual algorithms. Conformal inference techniques improves the performance of the prediction intervals.

Keywords Emergency department, Time series forecasting, Machine learning, Ensemble learning, Confor- mal inference

Participants: Margaux Zaffran.

Results: Adaptive probabilistic forecasting of French electricity spot prices 34

Electricity price forecasting (EPF) plays a major role for electricity companies as a fundamental entry for trading decisions or energy management operations. As electricity can not be stored, electricity prices are highly volatile which make EPF a particularly difficult task. This is all the more true when dramatic fortuitous events disrupt the markets. Trading and more generally energy management decisions require risk management tools which are based on probabilistic EPF (PEPF). In this challenging context, we argue in favor of the deployment of highly adaptive black-boxes strategies allowing to turn any forecasts into a robust adaptive predictive interval, such as conformal prediction and online aggregation, as a fundamental last layer of any operational pipeline. We propose to investigate a novel data set containing the French electricity spot prices during the turbulent 2020-2021 years, and build a new explanatory feature revealing high predictive power, namely the nuclear availability. Benchmarking state-of-the-art PEPF on this data set highlights the difficulty of choosing a given model, as they all behave very differently in practice, and none of them is reliable. However, we propose an adequate conformalization, coined Online Sequential Split Conformal Prediction (OSSCP-horizon), that improves the performances of PEPF methods, even in the most hazardous period of late 2021. Finally, we emphasize that combining it with online aggregation significantly outperforms any other approaches, and should be the preferred pipeline, as it provides trustworthy probabilistic forecasts.

Results: Impact of liquid sublingual immunotherapy on asthma onset and progression in patients with allergic rhinitis: a nationwide population-based study (EfficAPSI study)

Background: The only disease-modifying treatment currently available for allergic rhinitis (AR) is allergen immunotherapy (AIT). The main objective of the EfficAPSI real-world study (RWS) was to evaluate the impact of liquid sublingual immunotherapy (SLIT-liquid) on asthma onset and evolution in AR patients.

Methods:

An analysis with propensity score weighting was performed using the EfficAPSI cohort, comparing patients dispensed SLIT-liquid with patients dispensed AR symptomatic medication with no history of AIT (controls). Index date corresponded to the first dispensation of either treatment. The sensitive definition of asthma event considered the first asthma drug dispensation, hospitalization or long-term disease (LTD) for asthma, the specific one omitted drug dispensation and the combined one considered omalizumab or three ICS ± LABA dispensation, hospitalization or LTD. In patients with pre-existing asthma, the GINA treatment step-up evolution was analyzed.

Findings: In this cohort including 112,492 SLIT-liquid and 333,082 controls, SLIT-liquid exposure was associated with a significant lower risk of asthma onset. Exposure to SLIT was associated with a one-third reduction in GINA step-up, irrespective of baseline treatment steps

Interpretation: In this national RWS with the largest number of person-years of follow-up to date in the field of AIT, SLIT-liquid was associated with a significant reduction in the risk of asthma onset or worsening. The use of three definitions (sensitive or specific) and GINA step-up reinforced the rigorous methodology, substantiating SLIT-liquid evidence as a causal treatment option for patients with respiratory allergies.

Participants: Julie Josse.

Results: Pilot deployment of a machine-learning enhanced prediction of need for hemorrhage resuscitation after trauma - the ShockMatrix pilot study 14

Importance: Decision-making in trauma patients remains challenging and often results in deviation from guidelines. Machine-Learning (ML) enhanced decision-support could improve hemorrhage resuscitation.

Aim: To develop a ML enhanced decision support tool to predict Need for Hemorrhage Resuscitation (NHR) (part I) and test the collection of the predictor variables in real time in a smartphone app (part II).

Design, setting, and participants: Development of a ML model from a registry to predict NHR relying exclusively on prehospital predictors. Several models and imputation techniques were tested. We also assess the feasibility to collect the predictors of the model in a customized smartphone app during prealert and generate a prediction in four level-1 trauma centers to compare the predictions to the gestalt of the trauma leader.

Main outcomes and measures: Part 1: Model output was NHR defined by 1) at least one RBC transfusion in resuscitation, 2) transfusion$\ge$4 RBC within 6 h, 3) any hemorrhage control procedure within 6 h or 4) death from hemorrhage within 24 h. The performance metric was the F4-score and compared to reference scores (RED FLAG, ABC). In part 2, the model and clinician prediction were compared with Likelihood Ratios (LR).

Results: From 36,325 eligible patients in the registry (Nov 2010—May 2022), 28,614 were included in the model development (Part 1). Median age was 36 [25-52], median ISS 13 [5-22], 3249/28614 (11%) corresponded to the definition of NHR. A XGBoost model with nine prehospital variables generated the best predictive performance for NHR according to the F4-score with a score of 0.76 [0.73-0.78]. Over a 3-month period (Aug-Oct 2022), 139 of 391 eligible patients were included in part II (38.5%), 22/139 with NHR. Clinician satisfaction was high, no workflow disruption observed and LRs comparable between the model and the clinicians.

Conclusions and relevance: The ShockMatrix pilot study developed a simple ML-enhanced NHR prediction tool demonstrating a comparable performance to clinical reference scores and clinicians. Collecting the predictor variables in real-time on prealert was feasible and caused no workflow disruption.

10.1.1 Visits of international scientists

10.1.2 Visits to international teams

10.2.1 PEPR Digital Health

The "PEPR Santé Numérique", launched in June 2023 as part of the Plan Innovation Santé 2030, is a major initiative in the "Digital Health" acceleration strategy with a program dedicated to stimulating scientific research in this field.

PreMeDICaL is involved in three projects that have been lauched:

• SMATCH "Statistical and AI Methods for the Challenges of Modern Clinical Trials in Digital Health" - Julie Josse , Pascal Demoly
  • New clinical trial methods and designs based on animal-to-human, research-based disease models,
  • Enriching clinical trials with multi-source, multi-dimensional ancillary data,
  • Next-generation designs for clinical evaluation of digital medical devices based on AI algorithms,
  • Regulation, feasibility and dissemination of clinical trials
• Digital Pharmacological Twins "Multi-scale and longitudinal data modelling in pharmacology: toward digital pharmacological twins" - Julie Josse
• Secure, safe and fair machine learning for healthcare - Aurélien Bellet

10.2.2 PEPR Cybersecurity

PreMeDICaL is involved in project IPoP (Interdisciplinary Project on Privacy) - Aurélien Bellet . The objectives of this project are to study the threats on privacy that have been introduced by these new services, and to conceive theoretical and technical privacy-preserving solutions that are compatible with French and European regulations, that preserve the quality of experience of the users. These solutions will be deployed and assessed, both on the technological and legal sides, and on their societal acceptability. In order to achieve these objectives, we adopt an interdisciplinary approach, bringing together many diverse fields: computer science, technology, engineering, social sciences, economy and law.

The project's scientific program focuses on new forms of personal information collection, on the learning of Artificial Intelligence (AI) models that preserve the confidentiality of personal information used, on data anonymization techniques, on securing personal data management systems, on differential privacy, on personal data legal protection and compliance, and all the associated societal and ethical considerations. This unifying interdisciplinary research program brings together internationally recognized research teams (from universities, engineering schools and institutions) working on privacy, and the French Data Protection Authority (CNIL).

This holistic vision of the issues linked to personal data protection will on one hand let us propose solutions to the scientific and technological challenges and, on the other hand, help us confront these solutions in many different ways in the context of interdisciplinary collaborations, thus leading to recommendations and proposals in the field of regulations or legal frameworks. This comprehensive consideration of all the issues aims at encouraging the adoption and acceptability of the solutions proposed by all stakeholders, legislators, data controllers, data processors, solution designers, developers all the way to end-users.

10.2.3 Inria Challenge FedMalin

Aurélien Bellet leads FedMalin. FedMalin is a research project that spans 11 Inria research teams and aims to push Federated Learning (FL) research and concrete use-cases through a multidisciplinary consortium involving expertise in ML, distributed systems, privacy and security, networks, and medicine. We propose to address a number of challenges that arise when FL is deployed over the Internet, including privacy & fairness, energy consumption, personalization, and location/time dependencies. FedMalin will also contribute to the development of open-source tools for FL experimentation and real-world deployments, and use them for concrete applications in medicine and crowdsensing.

The FedMalin Inria Challenge is supported by Groupe La Poste, sponsor of the Inria Foundation.

10.2.4 ANR JCJC PRIDE

Aurélien Bellet leads PRIDE, a JCJC ANR project on privacy-preserving decentralized machine learning. The goal of PRIDE is to develop theoretical and algorithmic tools that enable differentially-private ML methods operating on decentralized datasets, through three complementary objectives:

• Prove that decentralized learning protocols naturally amplify DP guarantees;
• Propose algorithms at the intersection of decentralized ML and secure multi-party computation;
• Design data-adaptive communication schemes to speed up the convergence on heterogeneous datasets.

10.2.5 Allergen-Chip-Challenge

The challenge L'allergen-chip-challenge aimed at creating a national dataset for artificial intelligence-assisted allergy diagnosis using semantic attributes and allergen multiplex technology. The challenge was supported by the Health Data Hub in collaboration with the company Trustee - Pascal Demoly

Three follow-up projects:

• grant PNRIA 2023 with Olivier Saut
• AAP MESSIDORE 2024 submitted, Pascal Demoly and Julie Josse lead one research axis
• Team retreat with Pascal Demoly and Julie Josse Julien Goret on Determination of molecular allergen profiles and links with respiratory and food allergies

10.2.6 Grant from the National Interministerial Road Safety Observatory

Julie Josse - In collaboration with Traumabase. Grant for the SPOTE project (Specificities of Populations and Impact of Territories) aimed at studying the intra-hospital outcome of victims of road accidents treated, in critical care, in France, between 2013 and 2027.

10.2.7 Grant from PHRC

Nicolas Molinari leads 3 work packages

• Evaluation of early venous stenting treatment of patients with newly diagnosed idiopathic intracranial hypertension
• Evaluation of venous stenting treatment of patients with idiopathic intracranial hypertension to pursue acetazolamide withdrawal
• REVERT - Reversing airway remodeling with Tezepelumab

10.2.8 Grant from Institut Exposum Doctoral Nexus

Nicolas Molinari obtained a grant from ExposUM Nexus 2024 Doctoral Nexus for Phd students on "Modeling suicide risk," principal investigator of the axis (196,000 Euros).

10.2.9 Grant from Directorate General for Healthcare Services (DGOS)

Nicolas Molinari obtained a grant from the Health Data and Applications (DAtAE)" call for projects launched by the Directorate General for Healthcare Services (DGOS) and operated by the Health Data Hub for the APPCMMAF study to improve the care of patients on continuous positive airway pressure (CPAP), principal investigator (269,648 Euros).

UM Envi-H

Initiative by the University of Montpellier.

The University of Montpellier, with the support of the Regional Health Agency of Occitanie, is launching an innovative project in the field of environmental health education: the creation of a Small Private Online Course (SPOC) dedicated to environmental health (EH) for primary care. This project is part of Axis 1, "Inform, educate, and train in environmental health," of the Regional Environmental Health Plan for Occitanie (PRSE4 Occitanie 2023-2028), which "aims to provide professionals, local authorities, and citizens with the knowledge and skills needed to act on environmental and health issues."

In collaboration with the Hérault Primary Health Insurance Fund and the University Department of General Medicine, this SPOC will be a hybrid training program combining online modules with in-person sessions.

Available from early 2026, it aims to develop EH skills for learners in both continuing and initial education. It is primarily intended for coordinators of coordinated healthcare structures (Territorial Professional Health Communities - CPTS / Multidisciplinary Health Centers - MSP), as well as for students in related fields.

This program will focus on enhancing the EH competencies of participants through a hybrid format combining online and in-person learning.

Participants: Pascal Demoly, Nicolas Molinari, Julie Josse.

ComexIA Health Occitanie

Members of the steering committee for the Occitanie region's key challenge "AI for health": preparation of the call for proposals (12 co-financed PhD positions), selection of applications, dossier follow-up, and management of a 1.2M Euros budget.

Other local Projects the team is part of: Muse, eDOL, expos-UM, viA-UM, Fondation One Science Montpellier.

11.1.1 Scientific events: organisation

• Aurélien Bellet co-organizes the Federated Learning One World webinar (1100+ registered attendees) since May 2020.
• Aurélien Bellet : membre du comité scientifique des 55ièmes Journées de Statistique (JDS 2024)
• Margaux Zaffran : Journée Young Statisticians and Probabilists 12th YSP, Institut Henri Poincaré, Paris, Jan 2024.
• Margaux Zaffran , Charlotte Voinot : Recontres avec les conférencier.e.s invité.e.s et acteur.rice.s de la SFdS Rencontres JdS, Bordeaux, France, May 2024.
• Margaux Zaffran , Charlotte Voinot : Déjeuners scientifiques JdS, Bordeaux, France, May 2024.
• Margaux Zaffran : Mathematical Statistics Day, Paris, France
• Nicolas Molinari : Chair of the session "Explicability and causal inference: new ways of using data" at the 1st Biotherapies & AI in Occitanie workshop, October 2024.

11.1.2 Scientific events: selection

11.1.3 Journal

11.1.4 Invited talks

• Pascal Demoly : Futurapolis Santé, "Exposome : la chasse aux ennemis de nos poumons est ouverte", Oct. 2024.
• Pascal Demoly : Congrès Francophone d'Allergologie, Apr. 2024.
• Pascal Demoly : Journée annuelle de l’Institut ExposUM, Nov. 2024.
• Julie Josse : Bernoulli-IMS 11th World Congress in Probability and Statistics 2024, Talk in session on missing values.
• Julie Josse : Symposium on Causality (Panel), Sept. 2024, Florence.
• Julie Josse : 50 ans du CMAP, Centre de Mathématiques Appliquées de l'Ecole Polytechnique, Sept. 2024, Paris.
• Julie Josse : 2nd Global Symposium of Research Methodology Innovation in Trauma and Emergency Care, May 2024, Columbus Ohio.
• Julie Josse : European Conference of Causal Inference (Eurocim), Apr. 2024., Copenhague.
• Julie Josse : NIH Biostatistics Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, February 2024
• Aurélien Bellet : IMS International Conference on Statistics and Data Science (ICSDS), Nice.
• Aurélien Bellet : 3rd Workshop on Principles of Distributed Learning, Paris.
• Aurélien Bellet : Workshop on AI Auditing, Paris.
• Aurélien Bellet : Learning and Optimization in Côte d'Azur Workshop, Sophia-Antipolis.
• Aurélien Bellet : Inserm Workshop on Massive Genomic Data: Statistical and Bioinformatic Advances, Bordeaux.
• Aurélien Bellet : Privacy Alpine Seminar (Privaski), Corrençon en Vercors.
• Aurélien Bellet : L'Oréal, virtual.
• Aurélien Bellet : Owkin, virtual.
• Charlotte Voinot : Talk on introduction to causal inference, Stat4Plant Seminar INRAE, Jouy en josas, March 2024
• Charlotte Voinot : Talk on Causal survival analysis, MECOSA (Methodological and Computational Advances in Survival Analysis), Inria Paris, November 2024
• Remi Khellaf : Talk on introduction to causal inference, Stat4Plant Seminar INRAE, Jouy en josas, March 2024
• Jeffrey Näf : Talk on Missing Values icsds2024, Nice, December 2024
• Pan Zhao : IMS Asia-Pacific Rim Meeting. Invited Talk. January 4 - 7, 2024, Melbourne, Australia.
• Ioan Tudor Cebere : Talks on privacy auditing at Microsoft Research, Google Deepmind, Harvard and University of Toronto.
• Nicolas Molinari Invited speaker at the workshop "AI for bronchial diseases," December 13, 2024.

11.1.5 Leadership within the scientific community

• Pascal Demoly : full member of the Academy of Medicine, 1st division
• Pascal Demoly : Animation of the network e-allergies
• Pascal Demoly : president of the "French Society of Allergology"
• Pascal Demoly : WHO Collaborating Center for "Scientific Support for Classifications".
• Julie Josse is elected as a member of the R foundation and of the R Foundation Conference Committee. She is in the board of the French R committee (organization for coordinating R conferences "Les rencontres R") and involved in a task Forwards force on behalf of the R Foundation with the aim of increasing the participation of women and under-represented groups in the STEM community (founding member in 2015).
• Margaux Zaffran : President of "Groupe Jeunes Statisticien.ne.s"
• Charlotte Voinot : Treasurer of "Groupe Jeunes Statisticien.ne.s"
• Ioan Tudor Cebere : Privacy Attacks Workgroup Leadership for OpenDP.

11.1.6 Scientific expertise

• Julie Josse : Member of the Searching Committee for ENDOMIC, Inria. 2024.
• Julie Josse : Advisory Board of HORIZON EUROPE (HORIZON-HLTH-2022-TOOL-11-02), more-europa 2023-.
• Julie Josse and Nicolas Molinari : Comité scientifique et éthique du CHU de Montpellier. Dec 2023 -
• Julie Josse : Evaluation of research projects for funding agency or promotions for tenured Professor positions. Washington University; John Hopkins University; ANRT (PhD Cifre);
• Aurélien Bellet : Member of the CNIL-Inria Privacy Award committee
• Aurélien Bellet : ethics advisor for the European Strategy Forum on Research Infrastructures (ESFRI) project SLICES-PP
• Nicolas Molinari : president of the Institutional Review Board (IRB) of the Adène group
• Nicolas Molinari : Expert for DGOS, ANR, and several GIRCI (research project evaluations).
• Nicolas Molinari : Scientific Advisory Board of Nomics, "Make sleep medicine accessible".

11.1.7 Research administration

• Aurélien Bellet : member of the Operational Committee for the assessment of Legal and Ethical risks (COERLE).
• Julie Josse member of CSD (“Comité Suivi Doctoral") Inria
• Nicolas Molinari : elected member of "Commissions scientifique spécialisées"(CSS) 6 of INSERM
• Margaux Zaffran Elected member, parity and diversity committee, CMAP, École polytechnique.

11.2.1 Teaching

• Engineering School: M2 students, 40heqTD, Introduction to Probabilistic Graphical Models and Deep Generative Models, Master recherche specialité "Mathématiques Appliquées", M2 Mathématiques, Vision et Apprentissage (ENS Paris-Saclay), 1er semestre, 2024/2025 Rémi Khellaf
• Master: Institut de formation en masso-kinésithérapie, 9heqTD, statistics, Montpellier - Nicolas Molinari
• Master: Institut de formation en masso-kinésithérapie, head of the program, statistics, Montpellier - Nicolas Molinari
• Ecoles d'étiopathie, head of the program, statistics, Montpellier - Nicolas Molinari
• Master: EDSB « Epidémiologie, Données de Santé, Biostatistique », head of « Grands enjeux en santé » , Université de Montpellier - Pascal Demoly

11.2.2 Supervision

PhD students:

• Julie Josse : Supervision of Laura Fuentes Vincente (grant Montpellier) with Antoine Chambaz, Nov 2024 -
• Julie Josse : Supervision of Ahmed Boughdiri (grant Inria), Sep 2023 -
• Julie Josse and Aurélien Bellet : Supervision of Rémi Khellaf (grant Montpellier) with Erwan Scornet, Sep 2023 -
• Julie Josse : Supervision of Charlotte Voinot with Bernard Sebastien (grant Phd thesis Cifre Sanofi), Apr. 2023 -
• Julie Josse : Supervision of the medical doctor (MD) Tobias Gauss with Pierre Bouzat (MD), Feb. 2023 -
• Julie Josse and Nicolas Molinari : Supervision of the MD Maxime Fosset (grant Montpellier University, MUSE) with Boris Jung (MD), May 2022 -
• Julie Josse : Supervision of Margaux Zaffran (Cifre EDF) with Aymeric Dieuleveut, Yannig Goude and Olivier Ferron, Defended June 2024.
• Julie Josse : Supervision of Pan Zhao (grant MUSE) with Antoine Chambaz, Defended September 2024.
• Aurélien Bellet : Supervision of Jean-Rémy Conti with Stéphan Clémençon, October 2021 -
• Aurélien Bellet : Supervision of Edwige Cyffers (defended in December 2024)
• Aurélien Bellet : Supervision of Ioan Tudor Cebere , October 2022 -
• Aurélien Bellet : Supervision of Clément Pierquin with Marc Tommasi, June 2023 -
• Aurélien Bellet : Supervision of Brahim Erraji with Catuscia Palamidessi and Michael Perrot, September 2023 -
• Aurélien Bellet : Supervision of Thomas Boudou with Batiste Le Bars, October 2024 -
• Aurélien Bellet : Supervision of Ghita Fassy El Fehri , December 2024 -
• Nicolas Molinari : Supervision of Coutureau J., December 2024 -
• Nicolas Molinari : Supervision of Ibrahim S., October 2024.
• Nicolas Molinari : Supervision of Marie Felicia Beclin , defended December 2024.
• Pascal Demoly Supervision of of Ileana Ghiordanescu, defended on December 3, 2024, entitled "Mathematical Modeling of Drug Hypersensitivity Reactions - From Phenotyping to Endotyping."

Postdocs:

• Julie Josse : Mathieu Even, Oct. 2024 - .
• Julie Josse : Houssam Zenati, Dec. 2023 - Dec. 2024. Joint supervision with Bertrand Thirion and Judith Abecassis.
• Julie Josse : Herb Susmann, Sept. 2023 - 2024. Joint supervision with Antoine Chambaz. Current position: postdoc NYU Grossman School of Medicine
• Julie Josse : Jeffrey Naf, Feb. 2023 -
• Aurélien Bellet : Batiste Le Bars, until July 2024
• Aurélien Bellet : Mathieu Dagreou, Dec 2024 -

Masters:

• Nicolas Molinari : Supervisor of the Master 2 internship (EDSB) of J. Coutureau (100%), "Score to differentiate malignant non-mass lesions and benign breast cancer," defended in June 2024.
• Nicolas Molinari : Co-supervisor of the Master 2 internship (EDSB) of M. Meerun (50%), "Prediction of mortality in severe acute pancreatitis," defended in June 2024.
• Nicolas Molinari : Supervisor of the Master 2 internship (EDSB) of F. Kucharczak (100%), "Contribution of statistical variability quantification in the diagnosis of Parkinson's disease," defended in June 2024.

11.2.3 Juries

Member of PhD/HDR committees:

• Julie Josse : CSI Eugène Berta, under the supervision of Francis Bach and Michael Jordan. 2024 -
• Julie Josse : PhD defense committee of Alexis Ayme under the supervision of Erwan Scornet, Claire Boyer and Aymeric Dieuleveut. Oct 2024.
• Julie Josse : PhD defense committee of Noemie Simon Tillaux, under the supervision of Florence tubach. Nov 2024.
• Julie Josse : HDR defense committee of Emilie Devijver. Nov 2024.
• Julie Josse : PhD defense committee Floriane Jochum, under the supervision of Anne Sophie Hamy. Dec. 2024.
• Julie Josse : PhD defense committee (reviewer) of Sophia Yazzourh under the supervision of Nicolas Savy and Philippe Saint Pierre.
• Julie Josse : Habilitation of Boris Hejblum, May 2024
• Julie Josse : CSI Rémy Chapelle, supervised by Bruno Falissard, Mohammed Sedki and Nicolas Vayatis. 2024 -
• Julie Josse : PhD defense committee of Armand Lacombe under the supervision of Michelle Sebag. Jan. 2024.
• Aurélien Bellet : Reviewer for the habilitation thesis (HDR) of Antoine Boutet. Dec. 2024.
• Aurélien Bellet : Reviewer for the PhD of Louis Leconte under the supervision of Eric Moulines, Lionel Trojman and Van Minh Nguyen. June 2024.
• Aurélien Bellet : Reviewer for the PhD of Mathieu Dagréou under the supervision of Samuel Vaiter and Thomas Moreau. Oct. 2024.
• Aurélien Bellet : PhD defense committee of Marie Garin under the supervision of Nicolas Vayatis. June 2024.
• Aurélien Bellet : PhD defense committee of Tanguy Lefort under the supervision of Joseph Salmon and Alexis Joly. Sep. 2024.
• Aurélien Bellet : PhD defense committee of Tuan-Anh Nguyen under the supervision of Denis Trystram and Kim Thang Nguyen. Oct. 2024.

Member of hiring committees:

• Julie Josse : Member of the committee Chaire de Professeur Junior, CBIO "Artificial Intelligence for Digital Health". Sep. 2024.
• Julie Josse : Member of the committee Chaire de Professeur Junior, ENS Lyon. June 2024.
• Julie Josse : Member of the committee Chaire de Professeur Junior, Statistics and Public Health - Inria Rennes. May 2024.
• Aurélien Bellet : Member of assistant professor recruiting committee - Université de Montpellier.

11.3.1 Specific official responsibilities in science outreach structures

• Julie Josse : Committee on Nomination for the Institute of Mathematical Statistics (IMS) to select one candidate for IMS President. 2024 -

11.3.2 Productions (articles, videos, podcasts, serious games, ...)

• Aurélien Bellet : article on federated learning for healthcare in Télécom Paris Alumni [link].
• Aurélien Bellet : Interview for LUM Magazine [link]
• Ioan Tudor Cebere : hosting OpenMined's Privacy Tech Talk Series on Youtube, see [link] and [link]

11.3.3 Participation in Live events

• Julie Josse : Table ronde sur l'IA - Infravia.
• Margaux Zaffran , Charlotte Voinot : Session groupe jeune JdS Sexisme ordinaire, violences sexistes et sexuelles, biais de genre. Quel est le constat aujourd'hui dans la recherche académique en France ?, Bordeaux, France, May 2024.
• Margaux Zaffran Volunteer, Séphora Berrebi Association. Participation to various masterclasses for high school girls.

International journals

• 11 articleL. S.Laura Sofia Cardelli, M.Mariarosaria Magaldi, A.Audrey Agullo, G.Gaetan Richard, E.Erika Nogue, P.Philippe Berdague, M.Michel Galiner, F.Frédéric Georger, F.François Picard, E.Elvira Prunet, N.Nicolas Molinari, A.Arnaud Bourdin, D.Dany Jaffuel and F.François Roubille. Sacubitril/valsartan has an underestimated impact on the right ventricle in patients with sleep-disordered breathing, especially central sleep apnoea syndrome.Archives of cardiovascular diseases2024, Online ahead of printIn press. HALDOI
• 12 articleB.Bénédicte Colnet, J.Julie Josse, G.Gaël Varoquaux and E.Erwan Scornet. Reweighting the RCT for generalization: finite sample error and variable selection.Journal of the Royal Statistical Society: Series A Statistics in SocietyMay 2024HALDOI
• 13 articleB.Bénédicte Colnet, I.Imke Mayer, G.Guanhua Chen, A.Awa Dieng, R.Ruohong Li, G.Gaël Varoquaux, J.-P.Jean-Philippe Vert, J.Julie Josse and S.Shu Yang. Causal inference methods for combining randomized trials and observational studies: a review.Statistical Science2024. In press. HALback to text
• 14 articleT.Tobias Gauss, J.-D.Jean-Denis Moyer, C.Clelia Colas, M.Manuel Pichon, N.Nathalie Delhaye, M.Marie Werner, V.Veronique Ramonda, T.Theophile Sempe, S.Sofiane Medjkoune, J.Julie Josse, A.Arthur James, A.Anatole Harrois, C.Caroline Jeantrelle, M.Mathieu Raux, J.Jean Pasqueron, C.Christophe Quesnel, A.Anne Godier, M.Mathieu Boutonnet, D.Delphine Garrigue, A.Alexandre Bourgeois, B.Benjamin Bijok, J.Julien Pottecher, A.Alain Meyer, P.Pierluigi Banco, E.Etienne Montalescau, E.Eric Meaudre, J.-L.Jean-Luc Hanouz, V.Valentin Lefrancois, G.Gérard Audibert, M.Marc Leone, E.Emmanuelle Hammad, G.Gary Duclos, T.Thierry Floch, T.Thomas Geeraerts, F.Fanny Bounes, J. B.Jean Baptiste Bouillon, B.Benjamin Rieu, S.Sébastien Gettes, N.Nouchan Mellati, L.Leslie Dussau, E.Elisabeth Gaertner, B.Benjamin Popoff, T.Thomas Clavier, P.Perrine Lepêtre, M.Marion Scotto, J.Julie Rotival, L.Loan Malec, C.Claire Jaillette, P.Pierre Gosset, C.Clément Collard, J.Jean Pujo, H.Hatem Kallel, A.Alexis Fremery, N.Nicolas Higel, M.Mathieu Willig, B.Benjamin Cohen, P. S.Paer Selim Abback, S.Samuel Gay, E.Etienne Escudier and R.Romain Mermillod Blondin. Pilot deployment of a machine-learning enhanced prediction of need for hemorrhage resuscitation after trauma – the ShockMatrix pilot study.BMC Medical Informatics and Decision Making241October 2024, 315HALDOIback to text
• 15 articleD.D Jaffuel, E.E Serrano, C.C Leroyer, A.A Chartier and P.P Demoly. SQ HDM sublingual immunotherapy tablet for the treatment of HDM allergic rhinitis and asthma improves subjective sleepiness and insomnia: an exploratory analysis of the real-life CARIOCA study.Journal of Investigational Allergology and Clinical Immunology3452024HALDOI
• 16 articleJ.Julie Josse, J. M.Jacob M. Chen, N.Nicolas Prost, G.Gaël Varoquaux and E.Erwan Scornet. On the consistency of supervised learning with missing values.Statistical Papers659March 2024, 5447-5479HALDOI
• 17 articleH.Holly Pan, D.Debbie Jarvis, J.James Potts, L.Lidia Casas, D.Dennis Nowak, J.Joachim Heinrich, J. G.Judith Garcia Aymerich, I.Isabel Urrutia, J.Jesus Martinez-Moratalla, J.-A.Jose-Antonio Gullon, A.Antonio Pereira-Vega, C.Chantal Raherison, S.Sebastien Chanoine, P.P Demoly, B.Benedicte Leynaert, T.Thorarinn Gislason, N.Nicole Probst, M. J.Michael J Abramson, R.Rain Jogi, D.Dan Norback, T.Torben Sigsgaard, M.Mario Olivieri, C.Cecilie Svanes and E.Elaine Fuertes. Gas cooking indoors and respiratory symptoms in the ECRHS cohort.International Journal of Hygiene and Environmental Health256March 2024, 114310HALDOI
• 18 articleA.Aude Sportisse, M.Matthieu Marbac, F.Fabien Laporte, G.Gilles Celeux, C.Claire Boyer, J.Julie Josse and C.Christophe Biernacki. Model-based Clustering with Missing Not At Random Data.Statistics and ComputingJune 2024HALDOI
• 19 articleJ.Jean‐baptiste Woillard, C.Clément Benoist, A.Alexandre Destere, M.Marc Labriffe, G.Giulia Marchello, J.Julie Josse and P.Pierre Marquet. To be or not to be, when synthetic data meet clinical pharmacology: A focused study on pharmacogenetics.CPT: Pharmacometrics and Systems PharmacologySeptember 2024, Online ahead of printHALDOI

International peer-reviewed conferences

• 20 inproceedingsC.Clément Bénard, J.Jeffrey Naf and J.Julie Josse. MMD-based Variable Importance for Distributional Random Forest.Proceedings of Machine Learning ResearchAISTATS 2024 - The 27th International Conference on Artificial Intelligence and StatisticsPMLR-238Volume 238: International Conference on Artificial Intelligence and Statistics, 2-4 May 2024, Palau de Congressos, Valencia, SpainValence, Spain2024, 1324-1332HAL
• 21 inproceedingsL.Louis Béthune, T.Thomas Massena, T.Thibaut Boissin, Y.Yannick Prudent, C.Corentin Friedrich, F.Franck Mamalet, A.Aurélien Bellet, M.Mathieu Serrurier, D.David Vigouroux and C.Corentin Friedrich. DP-SGD Without Clipping: The Lipschitz Neural Network Way.ICLR 2024 - 12th International Conference on Learning RepresentationsVienna (Austria), Austria2024HALback to text
• 22 inproceedingsE.Edwige Cyffers, A.Aurélien Bellet and J.Jalaj Upadhyay. Differentially Private Decentralized Learning with Random Walks.ICML 2024 - Forty-first International Conference on Machine LearningVienne (Autriche), AustriaarXiv2024HALDOIback to text
• 23 inproceedingsM.Mathieu Even, L.Luca Ganassali, J.Jakob Maier and L.Laurent Massoulié. Aligning Embeddings and Geometric Random Graphs: Informational Results and Computational Approaches for the Procrustes-Wasserstein Problem.NeurIPS 2024 - 38th Conference on Neural Information Processing SystemsVancouver (BC), CanadaDecember 2024HAL
• 24 inproceedingsH.Hadrien Hendrikx, P.Paul Mangold and A.Aurélien Bellet. The Relative Gaussian Mechanism and its Application to Private Gradient Descent.PMLRAISTATS 2024 - 27th International Conference on Artificial Intelligence and Statistics238Valencia, SpainAugust 2024, 3079-3087HALback to text
• 25 inproceedingsB.Batiste Le Bars, A.Aurélien Bellet, M.Marc Tommasi, K.Kevin Scaman and G.Giovanni Neglia. Improved Stability and Generalization Guarantees of the Decentralized SGD Algorithm.ICML 2024 - The Forty-first International Conference on Machine LearningVienne, AustriaJuly 2024HALback to text
• 26 inproceedingsA. E.Abdellah El Mrini, E.Edwige Cyffers and A.Aurélien Bellet. Privacy Attacks in Decentralized Learning.ICML 2024 - Forty-first International Conference on Machine LearningVienne (Austria), AustriaarXiv2024HALDOIback to text
• 27 inproceedingsC.Clément Pierquin, A.Aurélien Bellet, M.Marc Tommasi and M.Matthieu Boussard. Rényi Pufferfish Privacy: General Additive Noise Mechanisms and Privacy Amplification by Iteration via Shift Reduction Lemmas.International Conference on Machine Learning (ICML 2024)Vienna (Austria), Austria2024HALback to text
• 28 inproceedingsT.T Seoudi, D.D Ayache, O.O BENABBAD, F.F PAGES, J.J Charensol, M.M Bahriz, N.Nicolas Molinari, F.Fares Gouzi and A.Aurore Vicet. Breath analysis by quartz enhanced photoacoustic spectroscopy: A clinical study.FLAIR 2024 - Field Laser Applications in Industry and Research 2024Assise, ItalySeptember 2024HAL
• 29 inproceedingsA. S.Ali Shahin Shamsabadi, G.Gefei Tan, T. I.Tudor Ioan Cebere, A.Aurélien Bellet, H.Hamed Haddadi, N.Nicolas Papernot, X.Xiao Wang and A.Adrian Weller. Confidential-DPproof: Confidential Proof of Differentially Private Training.ICLR 2024 - 12th International Conference on Learning RepresentationsVienna (Austria), Austria2024HALback to text
• 30 inproceedingsP.Pan Zhao, A.Antoine Chambaz, J.Julie Josse and S.Shu Yang. Positivity-free Policy Learning with Observational Data.Proceedings of Machine Learning ResearchAISTATS 2024 - The 27th International Conference on Artificial Intelligence and StatisticsPMLR-238Volume 238: International Conference on Artificial Intelligence and Statistics, 2-4 May 2024, Palau de Congressos, Valencia, SpainValence, Spain2024, 1918-1926HAL

Conferences without proceedings

• 31 inproceedingsT.Tarek Seoudi, D.Diba Ayache, O.Oifa Benabbad, J.Julien Charensol, F.Fanny Pagès, E.Eric Rosenkrantz, N.Nicolas Molinari, M.Michaël Bahriz, F.Fares Gouzi and A.Aurore Vicet. Photoacoustic sensing based on resonant mechanical transducers: application to diagnosis in breath.ICPPP 2024 - 22st International Conference on Photoacoustic and Photothermal PhenomenaCoimbra (Portugal), PortugalJuly 2024HAL

Reports & preprints

• 32 miscA.Ahmed Boughdiri, J.Julie Josse and E.Erwan Scornet. Quantifying Treatment Effects: Estimating Risk Ratios in Causal Inference.October 2024HAL
• 33 miscT.Tudor Cebere, A.Aurélien Bellet and N.Nicolas Papernot. Tighter Privacy Auditing of DP-SGD in the Hidden State Threat Model.October 2024HALback to text
• 34 miscG.Grégoire Dutot, M.Margaux Zaffran, O.Olivier Féron and Y.Yannig Goude. Adaptive probabilistic forecasting of French electricity spot prices.May 2024HALback to text
• 35 miscP.Pierre Humbert, B.Batiste Le Bars, A.Aurélien Bellet and S.Sylvain Arlot. Marginal and training-conditional guarantees in one-shot federated conformal prediction.May 2024HALback to text
• 36 miscR.Rémi Khellaf, A.Aurélien Bellet and J.Julie Josse. Federated Causal Inference: Multi-Studies ATE Estimation beyond Meta-Analysis.October 2024HALback to text
• 37 miscC. J.Christian Janos Lebeda and L.Lukas Retschmeier. The Correlated Gaussian Sparse Histogram Mechanism.December 2024HALback to text
• 38 miscC. J.Christian Janos Lebeda and J.Jakub Tětek. Testing Identity of Distributions under Kolmogorov Distance in Polylogarithmic Space.October 2024HAL
• 39 miscG.Gaurav Maheshwari, A.Aurélien Bellet, P.Pascal Denis and M.Mikaela Keller. Synthetic Data Generation for Intersectional Fairness by Leveraging Hierarchical Group Structure.May 2024HALback to text
• 40 miscJ.Jeffrey Näf, P.Patrick Bachmann and M.Markus Meierer. Customer Base Analysis in Non-Contractual Settings: A Model of Customer Attrition, Transactions, and Spending.September 2024HAL
• 41 misc J.Jeffrey Näf, E.Erwan Scornet and J.Julie Josse. What Is a Good Imputation Under MAR Missingness? January 2025 HAL back to text
• 42 miscJ.Jeffrey Näf and H.Herbert Susmann. Causal-DRF: Conditional Kernel Treatment Effect Estimation using Distributional Random Forest.November 2024HAL
• 43 miscA. S.Ali Shahin Shamsabadi, P.Peter Snyder, R.Ralph Giles, A.Aurélien Bellet and H.Hamed Haddadi. Nebula: Efficient, Private and Accurate Histogram Estimation.September 2024HALback to text
• 44 miscL.Lena Stempfle, A.Arthur James, J.Julie Josse, T.Tobias Gauss and F.Fredrik Johansson. Expert Study on Interpretable Machine Learning Models with Missing Data.2024HALDOI
• 45 miscH.Herbert Susmann, A.Antoine Chambaz, J.Julie Josse, M.Mathias Wargon, P.Philippe Aegerter and E.Emmanuel Bacry. Probabilistic Prediction of Arrivals and Hospitalizations in Emergency Departments in Île-de-France.April 2024HALback to text
• 46 miscC.Charlotte Voinot, C.Clément Berenfeld, I.Imke Mayer, B.Bernard Sebastien and J.Julie Josse. Causal survival analysis, Estimation of the Average Treatment Effect (ATE): Practical Recommendations.December 2024HAL
• 47 miscM.Margaux Zaffran, J.Julie Josse, Y.Yaniv Romano and A.Aymeric Dieuleveut. Predictive Uncertainty Quantification with Missing Covariates.May 2024HALback to text