François Portet

Sannara EK, François Portet, Philippe Lalanda

Human Activity Recognition (HAR) on mobile devices has been demonstrated to be possible using neural models trained on data collected from the device's inertial measurement units. These models have used Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTMs), Transformers or a combination of these to achieve state-of-the-art results with real-time performance. However, these approaches have not been extensively evaluated in real-world situations where the input data may be different from the training data. This paper highlights the issue of data heterogeneity in machine learning applications and how it can hinder their deployment in pervasive settings. To address this problem, we propose and publicly release the code of two sensor-wise Transformer architectures called HART and MobileHART for Human Activity Recognition Transformer. Our experiments on several publicly available datasets show that these HART architectures outperform previous architectures with fewer floating point operations and parameters than conventional Transformers. The results also show they are more robust to changes in mobile position or device brand and hence better suited for the heterogeneous environments encountered in real-life settings. Finally, the source code has been made publicly available.

Riccardo Presotto, Sannara Ek, Gabriele Civitarese et al.

The use of supervised learning for Human Activity Recognition (HAR) on mobile devices leads to strong classification performances. Such an approach, however, requires large amounts of labeled data, both for the initial training of the models and for their customization on specific clients (whose data often differ greatly from the training data). This is actually impractical to obtain due to the costs, intrusiveness, and time-consuming nature of data annotation. Moreover, even with the help of a significant amount of labeled data, model deployment on heterogeneous clients faces difficulties in generalizing well on unseen data. Other domains, like Computer Vision or Natural Language Processing, have proposed the notion of pre-trained models, leveraging large corpora, to reduce the need for annotated data and better manage heterogeneity. This promising approach has not been implemented in the HAR domain so far because of the lack of public datasets of sufficient size. In this paper, we propose a novel strategy to combine publicly available datasets with the goal of learning a generalized HAR model that can be fine-tuned using a limited amount of labeled data on an unseen target domain. Our experimental evaluation, which includes experimenting with different state-of-the-art neural network architectures, shows that combining public datasets can significantly reduce the number of labeled samples required to achieve satisfactory performance on an unseen target domain.

Thierry Desot, François Portet, Michel Vacher

Spoken Language Understanding (SLU) is a core task in most human-machine interaction systems. With the emergence of smart homes, smart phones and smart speakers, SLU has become a key technology for the industry. In a classical SLU approach, an Automatic Speech Recognition (ASR) module transcribes the speech signal into a textual representation from which a Natural Language Understanding (NLU) module extracts semantic information. Recently End-to-End SLU (E2E SLU) based on Deep Neural Networks has gained momentum since it benefits from the joint optimization of the ASR and the NLU parts, hence limiting the cascade of error effect of the pipeline architecture. However, little is known about the actual linguistic properties used by E2E models to predict concepts and intents from speech input. In this paper, we present a study identifying the signal features and other linguistic properties used by an E2E model to perform the SLU task. The study is carried out in the application domain of a smart home that has to handle non-English (here French) voice commands. The results show that a good E2E SLU performance does not always require a perfect ASR capability. Furthermore, the results show the superior capabilities of the E2E model in handling background noise and syntactic variation compared to the pipeline model. Finally, a finer-grained analysis suggests that the E2E model uses the pitch information of the input signal to identify voice command concepts. The results and methodology outlined in this paper provide a springboard for further analyses of E2E models in speech processing.

Anastasiia Usmanova, François Portet, Philippe Lalanda et al.

Federated Learning has been introduced as a new machine learning paradigm enhancing the use of local devices. At a server level, FL regularly aggregates models learned locally on distributed clients to obtain a more general model. Current solutions rely on the availability of large amounts of stored data at the client side in order to fine-tune the models sent by the server. Such setting is not realistic in mobile pervasive computing where data storage must be kept low and data characteristic can change dramatically. To account for this variability, a solution is to use the data regularly collected by the client to progressively adapt the received model. But such naive approach exposes clients to the well-known problem of catastrophic forgetting. To address this problem, we have defined a Federated Continual Learning approach which is mainly based on distillation. Our approach allows a better use of resources, eliminating the need to retrain from scratch at the arrival of new data and reducing memory usage by limiting the amount of data to be stored. This proposal has been evaluated in the Human Activity Recognition (HAR) domain and has shown to effectively reduce the catastrophic forgetting effect.

Yongxin Zhou, François Portet, Fabien Ringeval

Pre-trained language models have established the state-of-the-art on various natural language processing tasks, including dialogue summarization, which allows the reader to quickly access key information from long conversations in meetings, interviews or phone calls. However, such dialogues are still difficult to handle with current models because the spontaneity of the language involves expressions that are rarely present in the corpora used for pre-training the language models. Moreover, the vast majority of the work accomplished in this field has been focused on English. In this work, we present a study on the summarization of spontaneous oral dialogues in French using several language specific pre-trained models: BARThez, and BelGPT-2, as well as multilingual pre-trained models: mBART, mBARThez, and mT5. Experiments were performed on the DECODA (Call Center) dialogue corpus whose task is to generate abstractive synopses from call center conversations between a caller and one or several agents depending on the situation. Results show that the BARThez models offer the best performance far above the previous state-of-the-art on DECODA. We further discuss the limits of such pre-trained models and the challenges that must be addressed for summarizing spontaneous dialogues.

Sannara Ek, François Portet, Philippe Lalanda et al.

Pervasive computing promotes the integration of smart devices in our living spaces to develop services providing assistance to people. Such smart devices are increasingly relying on cloud-based Machine Learning, which raises questions in terms of security (data privacy), reliance (latency), and communication costs. In this context, Federated Learning (FL) has been introduced as a new machine learning paradigm enhancing the use of local devices. At the server level, FL aggregates models learned locally on distributed clients to obtain a more general model. In this way, no private data is sent over the network, and the communication cost is reduced. Unfortunately, however, the most popular federated learning algorithms have been shown not to be adapted to some highly heterogeneous pervasive computing environments. In this paper, we propose a new FL algorithm, termed FedDist, which can modify models (here, deep neural network) during training by identifying dissimilarities between neurons among the clients. This permits to account for clients' specificity without impairing generalization. FedDist evaluated with three state-of-the-art federated learning algorithms on three large heterogeneous mobile Human Activity Recognition datasets. Results have shown the ability of FedDist to adapt to heterogeneous data and the capability of FL to deal with asynchronous situations.

Sannara Ek, Romain Rombourg, François Portet et al.

Federated Learning is a new machine learning paradigm dealing with distributed model learning on independent devices. One of the many advantages of federated learning is that training data stay on devices (such as smartphones), and only learned models are shared with a centralized server. In the case of supervised learning, labeling is entrusted to the clients. However, acquiring such labels can be prohibitively expensive and error-prone for many tasks, such as human activity recognition. Hence, a wealth of data remains unlabelled and unexploited. Most existing federated learning approaches that focus mainly on supervised learning have mostly ignored this mass of unlabelled data. Furthermore, it is unclear whether standard federated Learning approaches are suited to self-supervised learning. The few studies that have dealt with the problem have limited themselves to the favorable situation of homogeneous datasets. This work lays the groundwork for a reference evaluation of federated Learning with Semi-Supervised Learning in a realistic setting. We show that standard lightweight autoencoder and standard Federated Averaging fail to learn a robust representation for Human Activity Recognition with several realistic heterogeneous datasets. These findings advocate for a more intensive research effort in Federated Self Supervised Learning to exploit the mass of heterogeneous unlabelled data present on mobile devices.

Anastasiia Usmanova, François Portet, Philippe Lalanda et al.

Federated Learning has been introduced as a new machine learning paradigm enhancing the use of local devices. At a server level, FL regularly aggregates models learned locally on distributed clients to obtain a more general model. In this way, no private data is sent over the network, and the communication cost is reduced. However, current solutions rely on the availability of large amounts of stored data at the client side in order to fine-tune the models sent by the server. Such setting is not realistic in mobile pervasive computing where data storage must be kept low and data characteristic (distribution) can change dramatically. To account for this variability, a solution is to use the data regularly collected by the client to progressively adapt the received model. But such naive approach exposes clients to the well-known problem of catastrophic forgetting. The purpose of this paper is to demonstrate this problem in the mobile human activity recognition context on smartphones.

Yongxin Zhou, Fabien Ringeval, François Portet

Automatic dialogue summarization is a well-established task with the goal of distilling the most crucial information from human conversations into concise textual summaries. However, most existing research has predominantly focused on summarizing factual information, neglecting the affective content, which can hold valuable insights for analyzing, monitoring, or facilitating human interactions. In this paper, we introduce and assess a set of measures PSentScore, aimed at quantifying the preservation of affective content in dialogue summaries. Our findings indicate that state-of-the-art summarization models do not preserve well the affective content within their summaries. Moreover, we demonstrate that a careful selection of the training set for dialogue samples can lead to improved preservation of affective content in the generated summaries, albeit with a minor reduction in content-related metrics.

Ali Can Kocabiyikoglu, François Portet, Prudence Gibert et al.

Spoken medical dialogue systems are increasingly attracting interest to enhance access to healthcare services and improve quality and traceability of patient care. In this paper, we focus on medical drug prescriptions acquired on smartphones through spoken dialogue. Such systems would facilitate the traceability of care and would free clinicians' time. However, there is a lack of speech corpora to develop such systems since most of the related corpora are in text form and in English. To facilitate the research and development of spoken medical dialogue systems, we present, to the best of our knowledge, the first spoken medical drug prescriptions corpus, named PxSLU. It contains 4 hours of transcribed and annotated dialogues of drug prescriptions in French acquired through an experiment with 55 participants experts and non-experts in prescriptions. We also present some experiments that demonstrate the interest of this corpus for the evaluation and development of medical dialogue systems.

Yongxin Zhou, Fabien Ringeval, François Portet

This study investigates the ability of GPT models (ChatGPT, GPT-4 and GPT-4o) to generate dialogue summaries that adhere to human guidelines. Our evaluation involved experimenting with various prompts to guide the models in complying with guidelines on two datasets: DialogSum (English social conversations) and DECODA (French call center interactions). Human evaluation, based on summarization guidelines, served as the primary assessment method, complemented by extensive quantitative and qualitative analyses. Our findings reveal a preference for GPT-generated summaries over those from task-specific pre-trained models and reference summaries, highlighting GPT models' ability to follow human guidelines despite occasionally producing longer outputs and exhibiting divergent lexical and structural alignment with references. The discrepancy between ROUGE, BERTScore, and human evaluation underscores the need for more reliable automatic evaluation metrics.

Marco Dinarelli, Marco Naguib, François Portet

Recent advances in spoken language understanding benefited from Self-Supervised models trained on large speech corpora. For French, the LeBenchmark project has made such models available and has led to impressive progress on several tasks including spoken language understanding. These advances have a non-negligible cost in terms of computation time and energy consumption. In this paper, we compare several learning strategies trying to reduce such cost while keeping competitive performance. At the same time we propose an extensive analysis where we measure the cost of our models in terms of training time and electric energy consumption, hopefully promoting a comprehensive evaluation procedure. The experiments are performed on the FSC and MEDIA corpora, and show that it is possible to reduce the learning cost while maintaining state-of-the-art performance and using SSL models.

Marco Naguib, François Portet, Marco Dinarelli

Recent advances in spoken language understanding benefited from Self-Supervised models trained on large speech corpora. For French, the LeBenchmark project has made such models available and has led to impressive progress on several tasks including spoken language understanding. These advances have a non-negligible cost in terms of computation time and energy consumption. In this paper, we compare several learning strategies aiming at reducing such cost while keeping competitive performances. The experiments are performed on the MEDIA corpus, and show that it is possible to reduce the learning cost while maintaining state-of-the-art performances.

Phuong-Hang Le, Valentin Pelloin, Arnault Chatelain et al.

We release Pantagruel models, a new family of self-supervised encoder models for French text and speech. Instead of predicting modality-tailored targets such as textual tokens or speech units, Pantagruel learns contextualized target representations in the feature space, allowing modality-specific encoders to capture linguistic and acoustic regularities more effectively. Separate models are pre-trained on large-scale French corpora, including Wikipedia, OSCAR and CroissantLLM for text, together with MultilingualLibriSpeech, LeBenchmark, and INA-100k for speech. INA-100k is a newly introduced 100,000-hour corpus of French audio derived from the archives of the Institut National de l'Audiovisuel (INA), the national repository of French radio and television broadcasts, providing highly diverse audio data. We evaluate Pantagruel across a broad range of downstream tasks spanning both modalities, including those from the standard French benchmarks such as FLUE or LeBenchmark. Across these tasks, Pantagruel models show competitive or superior performance compared to strong French baselines such as CamemBERT, FlauBERT, and LeBenchmark2.0, while maintaining a shared architecture that can seamlessly handle either speech or text inputs. These results confirm the effectiveness of feature-space self-supervised objectives for French representation learning and highlight Pantagruel as a robust foundation for multimodal speech-text understanding.

Kentaro Ueda, François Portet, Hirohiko Suwa et al.

While LLMs excel at general tasks, they struggle in specialized domains like finance, requiring diverse skills in domain knowledge, mathematical reasoning, and multilingual processing. Merging domain-specific Continual Pre-training (CPT) "experts" offers a practical alternative to costly and unstable multi-skill training. However, unlike established Supervised Fine-Tuning (SFT) model-based merging, CPT model merging remains largely unexplored. We address this gap by creating financial LLMs from experts in finance, math, and Japanese. We propose a three-stage evaluation focusing on knowledge recovery, complementarity, and emergence, and assess three merging methods (Task Arithmetic, TIES, and DARE-TIES) on a comprehensive financial benchmark curated from 18 tasks across 8 established datasets. Results show that merging an expert with its base model recovers general knowledge lost during CPT, while merging experts improves performance and can yield emergent cross-domain skills. Among the methods, Task Arithmetic performs strongly but is hyperparameter-sensitive, whereas TIES is more robust. Our findings also suggest that while model similarity correlates with merging success, emergent skills depend on more complex factors. This work presents the first foundational analysis of CPT model merging, establishing a principled framework and providing clear guidance for building multi-skill LLMs from existing assets.

Ali Can Kocabiyikoglu, François Portet, Jean-Marc Babouchkine et al.

Hospital information systems (HIS) have become an essential part of healthcare institutions and now incorporate prescribing support software. Prescription support software allows for structured information capture, which improves the safety, appropriateness and efficiency of prescriptions and reduces the number of adverse drug events (ADEs). However, such a system increases the amount of time physicians spend at a computer entering information instead of providing medical care. In addition, any new visiting clinician must learn to manage complex interfaces since each HIS has its own interfaces. In this paper, we present a natural language interface for e-prescribing software in the form of a spoken dialogue system accessible on a smartphone. This system allows prescribers to record their prescriptions verbally, a form of interaction closer to their usual practice. The system extracts the formal representation of the prescription ready to be checked by the prescribing software and uses the dialogue to request mandatory information, correct errors or warn of particular situations. Since, to the best of our knowledge, there is no existing voice-based prescription dialogue system, we present the system developed in a low-resource environment, focusing on dialogue modeling, semantic extraction and data augmentation. The system was evaluated in the wild with 55 participants. This evaluation showed that our system has an average prescription time of 66.15 seconds for physicians and 35.64 seconds for other experts, and a task success rate of 76\% for physicians and 72\% for other experts. All evaluation data were recorded and annotated to form PxCorpus, the first spoken drug prescription corpus that has been made fully available to the community (\url{https://doi.org/10.5281/zenodo.6524162}).

Felix Herron, Maja Hjuler, Solange Rossato et al.

Speech encoder models are known to model members of some speaker groups (SGs) better than others. However, there has been little work in establishing why this occurs on a technological level. To our knowledge, we present the first layerwise fairness analysis of pretrained self-supervised speech encoder models (S3Ms), probing each embedding layer for speaker identification (SID) automatic speech recognition (ASR). We find S3Ms produce embeddings biased against certain SGs for both tasks, starting at the very first latent layers. Furthermore, we find opposite patterns of layerwise bias for SID vs ASR for all models in our study: SID bias is minimized in layers that minimize overall SID error; on the other hand, ASR bias is maximized in layers that minimize overall ASR error. The inverse bias/error relationship for ASR is unaffected when probing S3Ms that are finetuned for ASR, suggesting SG-level bias is established during pretraining and is difficult to remove.

Maxime Méloux, Giada Dirupo, François Portet et al.

In a striking neuroscience study, the authors placed a dead salmon in an MRI scanner and showed it images of humans in social situations. Astonishingly, standard analyses of the time reported brain regions predictive of social emotions. The explanation, of course, was not supernatural cognition but a cautionary tale about misapplied statistical inference. In AI interpretability, reports of similar ''dead salmon'' artifacts abound: feature attribution, probing, sparse auto-encoding, and even causal analyses can produce plausible-looking explanations for randomly initialized neural networks. In this work, we examine this phenomenon and argue for a pragmatic statistical-causal reframing: explanations of computational systems should be treated as parameters of a (statistical) model, inferred from computational traces. This perspective goes beyond simply measuring statistical variability of explanations due to finite sampling of input data; interpretability methods become statistical estimators, and findings should be tested against explicit and meaningful alternative computational hypotheses, with uncertainty quantified with respect to the postulated statistical model. It also highlights important theoretical issues, such as the identifiability of common interpretability queries, which we argue is critical to understand the field's susceptibility to false discoveries, poor generalizability, and high variance. More broadly, situating interpretability within the standard toolkit of statistical inference opens promising avenues for future work aimed at turning AI interpretability into a pragmatic and rigorous science.

Felix Herron, Ange Richard, François Portet et al.

Many studies have shown automatic speech processing (ASR) systems have unequal performance across speakergroups (SG's). However, the manner in which such studies arrive at this conclusion is inconsistent. To pave the wayfor more reliable results in future studies, we lay out best practices for benchmarking ASR fairness based on literaturefrom machine learning fairness, social sciences, and speech science. We first describe the importance of preciselythe fairness hypothesis being interrogated, and tailoring fairness metrics to apply specifically to said hypothesis.We then examine several benchmarks used to rate ASR systems on fairness and discuss how their results can bemisconstrued without assiduous oversight into the intersections between SG's. We find that evaluating fairnessbased on single heterogeneous SG's, such as they are defined in fairness benchmarks, can lead to misidentifyingwhich SG's are actually being mistreated by ASR systems. We advocate for as fine-grained an analysis as possibleof the intersectionality of as many demographic variables as are available in the metadata of fairness corpora in orderto tease out such spurious correlations

Ange Richard, Laura Alonzo-Canul, François Portet

Quotation extraction is a widely useful task both from a sociological and from a Natural Language Processing perspective. However, very little data is available to study this task in languages other than English. In this paper, we present a manually annotated corpus of 1676 newswire texts in French for quotation extraction and source attribution. We first describe the composition of our corpus and the choices that were made in selecting the data. We then detail the annotation guidelines and annotation process, as well as a few statistics about the final corpus and the obtained balance between quote types (direct, indirect and mixed, which are particularly challenging). We end by detailing our inter-annotator agreement between the 8 annotators who worked on manual labelling, which is substantially high for such a difficult linguistic phenomenon.

Yongxin Zhou, Fabien Ringeval, François Portet

The growing demand for cognitive remediation therapy, combined with limited speech therapist availability, has accelerated the adoption of remote rehabilitation tools. These systems generate large volumes of interaction data that are difficult for clinicians to review efficiently. This paper investigates automated clinical report generation for avatar-guided, home-based cognitive remediation sessions in a low-resource setting with no reference reports. We present and compare two approaches: (1) a rule-based template system encoding speech therapy domain knowledge as explicit decision rules and validated templates, ensuring clinical reliability and traceability; and (2) a zero-shot LLM-based approach (GPT-4) aimed at more fluent and concise output. Both systems use identical pre-extracted, expert-validated structured variables, enabling a controlled factual comparison. Outputs were evaluated by eight speech therapists and final-year students using a nine-criterion questionnaire. Results reveal a clear trade-off between clinical reliability and linguistic quality. The template-based system scored higher on fluidity, coherence, and results presentation, while GPT-4 produced more concise output. Directional differences are consistent across evaluation dimensions, though no comparison reached statistical significance after correction, reflecting the scale constraints of expert clinical evaluation. Based on evaluator feedback, we derive eight design recommendations for clinical reporting systems in remote rehabilitation settings. More broadly, this work contributes a replicable methodology combining expert elicitation, taxonomy-driven generation, and multi-dimensional human evaluation for clinical NLG in low-resource settings, and illustrates how controlled comparisons can inform the responsible adoption of generative AI in healthcare.

Hippolyte Fournier, Sina Alisamir, Safaa Azzakhnini et al.

The integration of artificial intelligence (AI) into healthcare has advanced significantly, yet affect recognition remains a major challenge, particularly in AI-assisted interventions such as Computerized Cognitive Training (CCT). The THERADIA-WoZ corpus was developed to enable multimodal affect recognition in the context of AI-driven CCT, focusing on an older adult population. This study extends the corpus by introducing a dataset collected from young adults, allowing direct comparison of affect recognition models across age groups. Our objective was to assess whether multimodal models based on dimensions borrowed from appraisal theories outperform those based on categorical labels and to evaluate their generalisation power across age corpora. After comparing both corpora, models were trained and tested using within-corpus, cross-corpus, and mixed-corpus evaluation. Results revealed that appraisal dimensions consistently outperformed categorical labels across all conditions, demonstrating greater predictive accuracy and stability. Notably, categorical labels failed to generalise across age corpora, as performance dropped to chance levels in cross-corpus evaluation. In contrast, appraisal dimensions maintained predictive performance above chance, reinforcing their robustness for cross-age affect recognition. Furthermore, training on both corpora did not improve generalisation beyond within-corpus training. The findings support the theoretical and practical advantages of appraisal dimensions over categorical labels in affective computing. They also highlight the importance of multimodal fusion and deep learning representations for emotion modeling. To facilitate future research, we provide an API for researchers interested in time-continuous emotion prediction, offering valuable tools for behavioral sciences to enhance the measurement of emotional states in various experimental settings.

Xinhao Zhang, Xi Chen, François Portet et al.

Recent work has demonstrated the promise of orchestrating large language models (LLMs) within evolutionary and agentic optimization systems. However, the mechanisms driving these optimization gains remain poorly understood. In this work, we present a large-scale study of LLM-guided evolutionary search, collecting optimization trajectories for 15 LLMs across 8 tasks. Although zero-shot problem-solving ability correlates with final optimization outcomes, it explains only part of the variance: models with similar initial capability often induce dramatically different search trajectories and outcomes. By analyzing these trajectories, we find that strong LLM optimizers behave as local refiners, producing frequent incremental improvements while progressively localizing the search in semantic space. Conversely, weaker optimizers exhibit large semantic drift, with sporadic breakthroughs followed by stagnation. Notably, various measures of solution novelty do not predict final performance; novelty is beneficial only when the search remains sufficiently localized around high-performing regions of the solution space. Our results highlight the importance of trajectory analysis for understanding and improving LLM-based optimization systems and provide actionable insights for their design and training.

Maxime Méloux, Silviu Maniu, François Portet et al.

As AI systems are used in high-stakes applications, ensuring interpretability is crucial. Mechanistic Interpretability (MI) aims to reverse-engineer neural networks by extracting human-understandable algorithms to explain their behavior. This work examines a key question: for a given behavior, and under MI's criteria, does a unique explanation exist? Drawing on identifiability in statistics, where parameters are uniquely inferred under specific assumptions, we explore the identifiability of MI explanations. We identify two main MI strategies: (1) "where-then-what," which isolates a circuit replicating model behavior before interpreting it, and (2) "what-then-where," which starts with candidate algorithms and searches for neural activation subspaces implementing them, using causal alignment. We test both strategies on Boolean functions and small multi-layer perceptrons, fully enumerating candidate explanations. Our experiments reveal systematic non-identifiability: multiple circuits can replicate behavior, a circuit can have multiple interpretations, several algorithms can align with the network, and one algorithm can align with different subspaces. Is uniqueness necessary? A pragmatic approach may require only predictive and manipulability standards. If uniqueness is essential for understanding, stricter criteria may be needed. We also reference the inner interpretability framework, which validates explanations through multiple criteria. This work contributes to defining explanation standards in AI.

Felix Herron, Solange Rossato, Alexandre Allauzen et al.

Modern automatic speech recognition (ASR) systems have been observed to function better for certain speaker groups (SGs) than others, despite recent gains in overall performance. One potential impediment to progress towards fairer ASR is a more nuanced understanding of the types of modeling errors that speech encoder models make, and in particular the difference between the structure of embeddings for high-performance and low-performance SGs. This paper proposes a framework typifying two types of error that can occur in modeling phonemes in ASR systems: random error/high variance in phoneme embedding, vs systematic error/embedding bias. We find that training phoneme classification probes only on a single, typically disadvantaged SG, sometimes improves performance for that SG, which is evidence for the existence of SG-level bias in phoneme embeddings. On the other hand, we find that speakers and SGs with higher levels of phoneme variance are the same as those with worse phoneme prediction accuracy. We conclude that both types of error are present in phoneme embeddings and both are candidate causes for SG-level unfairness in ASR, though random error is likely a greater hindrance to fairness than systematic error. Furthermore, we find that finetuning encoder models using a fairness-enhancing algorithm (domain enhancing and adversarial training) changes neither the benefits of in-domain phoneme classification probe training, nor measured levels of random embedding error.

Maxime Méloux, François Portet, Maxime Peyrard

The development of trustworthy artificial intelligence requires moving beyond black-box performance metrics toward an understanding of models' internal computations. Mechanistic Interpretability (MI) aims to meet this need by identifying the algorithmic mechanisms underlying model behaviors. Yet, the scientific rigor of MI critically depends on the reliability of its findings. In this work, we argue that interpretability methods, such as circuit discovery, should be viewed as statistical estimators, subject to questions of variance and robustness. To illustrate this statistical framing, we present a systematic stability analysis of a state-of-the-art circuit discovery method: EAP-IG. We evaluate its variance and robustness through a comprehensive suite of controlled perturbations, including input resampling, prompt paraphrasing, hyperparameter variation, and injected noise within the causal analysis itself. Across a diverse set of models and tasks, our results demonstrate that EAP-IG exhibits high structural variance and sensitivity to hyperparameters, questioning the stability of its findings. Based on these results, we offer a set of best-practice recommendations for the field, advocating for the routine reporting of stability metrics to promote a more rigorous and statistically grounded science of interpretability.

Sannara Ek, Kaile Wang, François Portet et al.

Personalized Federated Learning (PFL) enables distributed training on edge devices, allowing models to collaboratively learn global patterns while tailoring their parameters to better fit each client's local data, all while preserving data privacy. However, PFL faces two key challenges in mobile systems: client drift, where heterogeneous data cause model divergence, and the overlooked need for client generalization, as the dynamic of mobile sensing demands adaptation beyond local environments. To overcome these limitations, we introduce Federated Alignment (FedAli), a prototype-based regularization technique that enhances inter-client alignment while strengthening the robustness of personalized adaptations. At its core, FedAli introduces the ALignment with Prototypes (ALP) layer, inspired by human memory, to enhance generalization by guiding inference embeddings toward personalized prototypes while reducing client drift through alignment with shared prototypes during training. By leveraging an optimal transport plan to compute prototype-embedding assignments, our approach allows pre-training the prototypes without any class labels to further accelerate convergence and improve performance. Our extensive experiments show that FedAli significantly enhances client generalization while preserving strong personalization in heterogeneous settings.

Ange Richard, Gilles Bastin, François Portet

In this article, we present {\it GenderedNews} (\url{https://gendered-news.imag.fr}), an online dashboard which gives weekly measures of gender imbalance in French online press. We use Natural Language Processing (NLP) methods to quantify gender inequalities in the media, in the wake of global projects like the Global Media Monitoring Project. Such projects are instrumental in highlighting gender imbalance in the media and its very slow evolution. However, their generalisation is limited by their sampling and cost in terms of time, data and staff. Automation allows us to offer complementary measures to quantify inequalities in gender representation. We understand representation as the presence and distribution of men and women mentioned and quoted in the news -- as opposed to representation as stereotypification. In this paper, we first review different means adopted by previous studies on gender inequality in the media : qualitative content analysis, quantitative content analysis and computational methods. We then detail the methods adopted by {\it GenderedNews} and the two metrics implemented: the masculinity rate of mentions and the proportion of men quoted in online news. We describe the data collected daily (seven main titles of French online news media) and the methodology behind our metrics, as well as a few visualisations. We finally propose to illustrate possible analysis of our data by conducting an in-depth observation of a sample of two months of our database.

Sannara Ek, François Portet, Philippe Lalanda et al.

Pervasive computing promotes the installation of connected devices in our living spaces in order to provide services. Two major developments have gained significant momentum recently: an advanced use of edge resources and the integration of machine learning techniques for engineering applications. This evolution raises major challenges, in particular related to the appropriate distribution of computing elements along an edge-to-cloud continuum. About this, Federated Learning has been recently proposed for distributed model training in the edge. The principle of this approach is to aggregate models learned on distributed clients in order to obtain a new, more general model. The resulting model is then redistributed to clients for further training. To date, the most popular federated learning algorithm uses coordinate-wise averaging of the model parameters for aggregation. However, it has been shown that this method is not adapted in heterogeneous environments where data is not identically and independently distributed (non-iid). This corresponds directly to some pervasive computing scenarios where heterogeneity of devices and users challenges machine learning with the double objective of generalization and personalization. In this paper, we propose a novel aggregation algorithm, termed FedDist, which is able to modify its model architecture (here, deep neural network) by identifying dissimilarities between specific neurons amongst the clients. This permits to account for clients' specificity without impairing generalization. Furthermore, we define a complete method to evaluate federated learning in a realistic way taking generalization and personalization into account. Using this method, FedDist is extensively tested and compared with three state-of-the-art federated learning algorithms on the pervasive domain of Human Activity Recognition with smartphones.

Ali Can Kocabiyikoglu, François Portet, Raheel Qader et al.

Drug prescriptions are essential information that must be encoded in electronic medical records. However, much of this information is hidden within free-text reports. This is why the medication extraction task has emerged. To date, most of the research effort has focused on small amount of data and has only recently considered deep learning methods. In this paper, we present an independent and comprehensive evaluation of state-of-the-art neural architectures on the I2B2 medical prescription extraction task both in the supervised and semi-supervised settings. The study shows the very competitive performance of simple DNN models on the task as well as the high interest of pre-trained models. Adapting the latter models on the I2B2 dataset enables to push medication extraction performances above the state-of-the-art. Finally, the study also confirms that semi-supervised techniques are promising to leverage large amounts of unlabeled data in particular in low resource setting when labeled data is too costly to acquire.

Anastasiia Usmanova, François Portet, Philippe Lalanda et al.

Federated Learning, a new machine learning paradigm enhancing the use of edge devices, is receiving a lot of attention in the pervasive community to support the development of smart services. Nevertheless, this approach still needs to be adapted to the specificity of the pervasive domain. In particular, issues related to continual learning need to be addressed. In this paper, we present a distillation-based approach dealing with catastrophic forgetting in federated learning scenario. Specifically, Human Activity Recognition tasks are used as a demonstration domain.

Raheel Qader, François Portet, Cyril Labbé

In Natural Language Generation (NLG), End-to-End (E2E) systems trained through deep learning have recently gained a strong interest. Such deep models need a large amount of carefully annotated data to reach satisfactory performance. However, acquiring such datasets for every new NLG application is a tedious and time-consuming task. In this paper, we propose a semi-supervised deep learning scheme that can learn from non-annotated data and annotated data when available. It uses an NLG and a Natural Language Understanding (NLU) sequence-to-sequence models which are learned jointly to compensate for the lack of annotation. Experiments on two benchmark datasets show that, with limited amount of annotated data, the method can achieve very competitive results while not using any pre-processing or re-scoring tricks. These findings open the way to the exploitation of non-annotated datasets which is the current bottleneck for the E2E NLG system development to new applications.

Alexis Brenon, François Portet, Michel Vacher

In a voice-controlled smart-home, a controller must respond not only to user's requests but also according to the interaction context. This paper describes Arcades, a system which uses deep reinforcement learning to extract context from a graphical representation of home automation system and to update continuously its behavior to the user's one. This system is robust to changes in the environment (sensor breakdown or addition) through its graphical representation (scale well) and the reinforcement mechanism (adapt well). The experiments on realistic data demonstrate that this method promises to reach long life context-aware control of smart-home.