INRIA Grenoble PhD Seminars

July 2026

Speakers :

• Luben Miguel Cruz Cabezas, PhD student at STATIFY team (Inria) : Towards Local and Epistemic Uncertainty Quantification with Conformal Prediction
  Google Scholar Website Slides
   Conformal Prediction (CP) is a distribution-free framework for uncertainty quantification: given any predictive model, treated as a black box, it produces prediction sets guaranteed to contain the true outcome with a user-specified probability, relying only on weak assumptions about the underlying data distribution, such as exchangeability. Despite these strong guarantees, standard CP methods have two important limitations. First, they typically produce prediction sets of constant size, failing to adapt to regions where the model is more or less uncertain. Second, they do not explicitly account for epistemic uncertainty — the uncertainty coming from limited data or an imperfect model — which can make prediction sets overconfident precisely where the model knows the least. This talk introduces the basics of CP and presents two of our contributions, each tackling one of these limitations: LOCART, which achieves local adaptivity by partitioning the feature space with a regression tree trained on conformity scores, and EPICSCORE, which enhances any conformal score by explicitly incorporating a Bayesian model of epistemic uncertainty. Together, they illustrate two complementary ways of making conformal prediction more informative in practice.


• Eduardo Steve Rodriguez Canales, PhD student at DANCE team (Inria) : Modeling and Control of Cycling Adoption Dynamics
  Google Scholar Slides
   The transition toward active transportation modes, particularly cycling, offers well-documented economic, environmental, and health benefits. However, mode choice is driven by a complex interplay between individual behavior, social influence, and the gradual evolution of urban infrastructure, making the design of effective public policies particularly challenging. In this presentation, I will introduce a dynamical systems framework — based on a nonlinear compartmental model — for modeling the population-level adoption of cycling, capturing behavioral transitions, social influence, and time-varying contextual factors. This model provides a natural basis for systems and control theory to design and analyze policy interventions operating over different time scales: feedback strategies can support short-term adaptation to uncertainties, while predictive optimization enables the planning of long-term investment policies that balance economic costs with sustained increases in cycling adoption. This talk will show how modeling, qualitative dynamical analysis, and control methods can be combined to better understand behavioral dynamics and to support robust, cost-effective, and implementable transportation policies.