This week

Mesoscopic stochastic systems at equilibrium are constrained by detailed balance and cannot sustain directed motion or perform work. It is only when they are driven out of equilibrium that they acquire the ability to act: transducing energy, generating forces, and exhibiting nontrivial responses.

In this talk, I present two experimental approaches to induce and exploit such nonequilibrium behavior in interacting many-body systems across micro- and macroscopic scales.

The first implements a Maxwell demon: by measuring the fluctuating state of a colloidal suspension and applying feedback, the system is driven out of equilibrium. This enables an information engine that extracts work from a single heat bath and transfers it to a probe particle.

The second system is a chiral active fluid composed of self-propelled macroscopic bristle-bots. Here, nonequilibrium arises from sustained energy injection at the level of individual propulsion. Because of the fluid’s chirality, a probe particle embedded in this medium exhibits an odd response to external forces, beyond the standard rheology of complex fluids.

Theoretical and applied research into privacy encompasses an incredibly broad swathe of differing approaches, emphass and aims. In a first part, we propose a novel framework for measuring privacy from a Bayesian game-theoretic perspective. This framework enables the creation of new, purpose-driven privacy definitions that are rigorously justified, while also allowing for the assessment of existing privacy guarantees through game theory. We show that pure and probabilistic differential privacy are special cases of our framework, and provide new interpretations of the post-processing inequality in this setting. Further, we demonstrate that privacy guarantees can be established for deterministic algorithms, which are overlooked by current privacy standards. In a second part, we introduce a new quantitative notion of privacy that is both contextual and specific. We argue that it provides a more meaningful notion of privacy than the widely utilised framework of differential privacy and a more explicit and rigorous formulation than what is commonly used in statistical disclosure theory. Our definition relies on concepts inherent to standard Bayesian decision theory, while departing from it in several important respects. In particular, the party controlling the release of sensitive information should make disclosure decisions from the prior viewpoint, rather than conditional on the data, even when the data is itself observed.

Joint work with James Bailie, Cameron Bell, Joshua Bon, Timothy Johnston, Antoine Luciano, and Judith Rousseau
https://arxiv.org/abs/2601.22945
https://arxiv.org/abs/2603.04199