Post-Doctoral Research Visit in Representation Learning and Clustering at Inria Rennes

A propos d'Inria

Inria est l'institut national de recherche dédié aux sciences et technologies du numérique. Il emploie 2600 personnes. Ses 215 équipes-projets agiles, en général communes avec des partenaires académiques, impliquent plus de 3900 scientifiques pour relever les défis du numérique, souvent à l'interface d'autres disciplines. L'institut fait appel à de nombreux talents dans plus d'une quarantaine de métiers différents. 900 personnels d'appui à la recherche et à l'innovation contribuent à faire émerger et grandir des projets scientifiques ou entrepreneuriaux qui impactent le monde. Inria travaille avec de nombreuses entreprises et a accompagné la création de plus de 200 start-up. L'institut s'eorce ainsi de répondre aux enjeux de la transformation numérique de la science, de la société et de l'économie.

Post-Doctoral Research Visit F/M Representation Learning and Clustering of Interacting Timed Systems Le descriptif de l'offre ci-dessous est en Anglais Type de contrat : CDD Niveau de diplôme exigé : Thèse ou équivalent Fonction : Post-Doctorant A propos du centre ou de la direction fonctionnelle The Inria Centre at Rennes University is one of Inria's nine centres and has more than thirty research teams. The Inria Centre is a major and recognized player in the field of digital sciences. It is at the heart of a rich R&D and innovation ecosystem: highly innovative PMEs, large industrial groups, competitiveness clusters, research and higher education players, laboratories of excellence, technological research institute, etc. Contexte et atouts du poste Scientific Context Recent advances in computer vision and sensor technologies enable the capture of rich, finely time-resolved information on individual behaviors. From video recordings and multimodal sensors, it is possible to detect behavioral state changes and estimate interactions between individuals within a group. These observations provide essential data for modeling both individual and collective dynamics. From an interpretability perspective, Timed Automata (TA) can be used as a symbolic representation framework for behavioral states and their temporal transitions. While learning a timed automaton for an individual over a single period is feasible, applying this approach to long-duration or repetitive behaviors involving multiple individuals quickly becomes uninformative and difficult to interpret. The objective of this postdoctoral position is therefore to define a typical behavior for each individual while preserving the temporal structures and interactions among individuals within a group. Keywords Neuro-Symbolic AI, Temporal Graph Representation Learning, Graph Variational Autoencoders, Clustering in Latent Space, Prototype Reconstruction, Behavioral Modeling. Objective The goal is to develop and integrate a methodological and software framework for the analysis of interacting timed systems. This will involve producing a prototype software platform combining latent representation learning, clustering, and behavior generation. This prototype will be particularly dedicated to analyzing behavioral dynamics and interactions among individuals, with a primary application to the study of animal welfare using multimodal data. Funding French PEPR Agriculture and ICT (WAIT4 project) Duration: 15-18 months, depending on the start date Starting date: As soon as possible from Spring 2026 Mission confiée Previous work developed within the team has already proposed a tool for learning timed automata from timed event sequence. The goal of this post-doctoral position is to develop a framework to analyze the behavior of multiple individuals using Timed Automata (TA). This framework will be scalable, interpretable, and generative, and will rely on embedding-based representations. The objective is to build compact representations of these behaviors while keeping important information about interactions and temporal constraints. Our aim is to build compact and expressive behavioral representations of TA while preserving interactions and temporal dynamics while preserving their interactions and temporal dynamics. A previous approach already enables the learning of TAs from behavioral data over long periods, which can be reused and further refined to effectively capture interactions between individuals within a group. Specifically, the objectives are: 1- Learning compact, latent representations of Timed Automata - Develop methods to embed individual TAs into a continuous latent space. - Explore approaches such as Graph Neural Network encoders, variational autoencoders (VAE), or alternative structured representation learning techniques. - Ensure that the latent representations capture both state transitions and temporal constraints preserving the behavioral semantics of each automaton. - Extend the framework to multi-automata systems while capturing the synchronizations between the other timed automata in interactions. 2- Clustering and prototype extraction - Perform clustering in the latent space to identify typical behavioral patterns among individuals and groups. - Construct prototype automata, providing interpretable summaries of typical individual and collective behaviors. - Apply clustering to detect anomalous or rare behaviors, enabling potential applications in monitoring or welfare assessment. 3- Evaluation and validation - Assess embeddings and clustering in terms of semantic preservation, behavioral interpretability, and interaction fidelity. - Compare alternative to identify the most effective approach for structured, temporal, interacting systems. - Evaluate on real datasets, including PEPR WAIT4 datasets on interactions between farm animals to estimate their welfare. Principales activités Expected Impact. This project is expected to make significant contributions at both the methodological and applied levels. - Research innovation: development of a novel approach for learning latent representations of symbolic, temporally structured systems such as Timed Automata, which has not been done before. - Interpretability: automata provide interpretable summaries of individual and collective behaviors. - Application to animal behavior: facilitates behavioral learning and monitoring of interactions in farm animals (e.g., using WAT4 datasets). - Broader applicability: framework can be applied to any interacting temporal system, beyond the motivating case study. - Bridging fields: combines neuro-symbolic AI, graph representation learning, and generative modeling in a unified approach. Compétences - The candidate should have strong programming skills and be comfortable working with latent representations, generative models, and structured formalisms. - The candidate should be able to collaborate with agronomy researchers on applications such as animal welfare - The ideal candidate is curious, independent, and motivated to contribute to research at the intersection of interpretable AI, representation learning, and interacting temporal systems. He has a strong motivation to contribution to an unexplored but promising domain. Avantages - Subsidized meals - Partial reimbursement of public transport costs - Leave: 7 weeks of annual leave + 10 extra days off due to RTT (statutory reduction in working hours) + possibility of exceptional leave (sick children, moving home, etc.) - Possibility of teleworking (after 6 months of employment) and flexible organization of working hours - Professional equipment available (videoconferencing, loan of computer equipment, etc.) - Social, cultural and sports events and activities - Access to vocational training Rémunération Monthly gross salary amounting to 2788 euros