Ben Kuznets-Speck, Northwestern University
Title: “Keeping score: inverting generative diffusion models for interpretable biophysical inference”
Abstract: Diffusion models excel at generating new samples from complex learned distributions, but they have seen limited use in inference and prediction tasks. We introduce a physics-inspired approach, Keeping SCORE, that transforms diffusion models into probabilistic engines for classification and regression. We calculate exact class likelihoods and quantify prediction confidence by measuring a thermodynamic dissipation (a function of the score) along noising trajectories under different class hypotheses. Our approach includes natural feature attributions that identify which input variables drive each decision, providing interpretability without modifying existing trained models. We test our framework across image recognition tasks (handwritten digits, natural photos), single-cell genomics (distinguishing cell identities, mapping gene perturbation effects), and molecular biophysics (predicting mutation impacts on protein folding energy). This connection between non-equilibrium statistical mechanics and modern AI approaches creates interpretable, uncertainty-aware predictions for biological discovery.