Prof. Philippe Lalanda

Université Grenoble Alpes, France

Abstract

Pervasive computing systems integrate sensors, connected devices, and software services into everyday environments in order to observe the physical world, interpret its context, and act accordingly. Such systems operate in highly dynamic and heterogeneous environments, where users, devices, and operating conditions continuously evolve. Autonomic computing has long provided principles and mechanisms to manage this complexity at runtime, through monitoring, analysis, decision making, and adaptation.

The growing use of machine learning is changing how pervasive services are built. By learning complex representations directly from sensor data, these services can go beyond what explicitly engineered context models and decision rules can capture. However, these capabilities come with a dependence on training data that creates new challenges after deployment. Changes in users, environments, or operating conditions can undermine model accuracy, while previously unseen situations can expose the limits of what models have learned. Maintaining reliable services therefore requires more than managing software components and computational resources: it also requires monitoring model behavior, assessing whether models remain suited to their operating context, and adapting them when necessary.

This talk explores how a MAPE-K loop (Monitor, Analyze, Plan, and Execute, supported by a shared Knowledge base) can support the autonomic management of machine learning models. The focus is on how to turn runtime observations into informed adaptation decisions: when to intervene, which model updates to perform, and how to assess their effects under explicit goals and constraints. We will discuss the techniques needed to support this loop, from runtime uncertainty estimation and change detection to model adaptation through continual learning. Applications in manufacturing fault detection and human activity recognition will illustrate how these techniques can be combined to reduce reliance on manual intervention while grounding adaptation decisions in sufficient evidence and accumulated knowledge (Lalanda et al. 2013; Balditsyn et al. 2026; Ek et al. 2025; Bredikhin et al. 2026).

Biography

Philippe Lalanda is a Professor at Université Grenoble Alpes (UGA), where he teaches software engineering and pervasive computing. He earned his PhD in artificial intelligence at Nancy University and subsequently worked on AI and robotics at Stanford University’s Knowledge Systems Laboratory (KSL). He then spent ten years in industry at Dassault Aviation, Thales, and Schneider Electric, holding positions as a software architect and R&D project leader.

His research focuses on autonomic computing and machine learning, particularly for pervasive computing systems. He has authored around 100 papers in international journals and conferences and supervised more than 20 PhD theses. He received an IBM Faculty Award in 2015.

References

• Balditsyn, Vladimir, Philippe Lalanda, and Stephanie Chollet. 2026. “Evaluating Uncertainty Quantification Methods for Flag-Based Monitoring of ML Systems in Pervasive Computing.” CCF Transactions on Pervasive Computing and Interaction, ahead of print. https://doi.org/10.1007/s42486-026-00271-x.
• Bredikhin, Aleksandr, Philippe Lalanda, and Germán Vega. 2026. “Efficient and Adaptive Human Activity Recognition via LLM Backbones.” 2026 IEEE International Conference on Smart Computing (SmartComp), 72–79. https://doi.org/10.1109/SmartComp69968.2026.00022.
• Ek, Sannara, Riccardo Presotto, Gabriele Civitarese, François Portet, Philippe Lalanda, and Claudio Bettini. 2025. “Comparing Self-Supervised Learning Techniques for Wearable Human Activity Recognition.” CCF Transactions on Pervasive Computing and Interaction 7 (3): 324–41.
• Lalanda, Philippe, Julie A. McCann, and Ada Diaconescu. 2013. Autonomic Computing - Principles, Design and Implementation. Undergraduate Topics in Computer Science. Springer. https://doi.org/10.1007/978-1-4471-5007-7.