Events

Date: 27 November 2025   Time: 13:00 - 14:00

Location: MB-503

Self-interacting jump processes are stochastic systems where transition rates depend on the process's own empirical occupation measure, introducing a feedback that breaks Markovianity. In this talk, I will present a large deviation framework for such systems. Specifically, I will derive the level-2.5 rate functional describing the joint fluctuations of occupation and flux, obtained via an exponential tilting construction extended to the non-Markovian setting. From this, I will show how memory modifies fluctuation bounds, leading to kinetic and thermodynamic uncertainty relations that generalise those of standard Markov processes. Time permitting, I will discuss one or two examples--a minimal two-state model and a collective exploration process inspired by ant dynamics--and outline a potential link to stochastic decision processes and reinforcement learning, illustrating how feedback reshapes fluctuation-dissipation trade-offs and can accelerate first-passage dynamics.