To adapt and thrive in uncertain environments, animals rely on their brains to estimate uncertainty across sensory and motor timescales. Sampling-based inference offers a robust theoretical framework for understanding probabilistic computation in neural circuits, but it remains unclear how spiking neural networks can implement these algorithms in a biologically plausible and computationally efficient way. Recent studies highlight the critical role of population geometry in enabling fast and accurate sampling. Moreover, novel non-reversible approaches, inspired by Piecewise Deterministic Markov Processes (PDMP) could further enhance efficiency and offer a biologically realistic path to scalable sampling. Exploring these ideas could provide new insights into neural computation and inspire advances in neuromorphic systems. The project is embedded in a interdisciplinary research collaboration bridging mathematics and computational neuroscience between Paul Masset (Mc Gill Uni), Manon Michel (Université Clermont-Auvergne) and Jacob Zavatone-Veth (Harvard Uni). The internship can be principally located either in Canada or France. This opportunity is open to Master’s students, including first-year students (Master 1). For Master 2 students, the internship will aim at laying the groundwork for an interdisciplinary PhD project centered on a quantitative analytical and numerical study of biologically-plausible sampling algorithms in spiking neural networks.