Imperfect Information Learning Team (Talk by Hanlin Yu, University of Helsinki)

RIKEN AIP's Imperfect Information Learning Team hosts a research talk by Hanlin Yu of the University of Helsinki, titled "Comparing Distributions and Representations: Geometry, Density Ratios, and Energy-Based Models." The session frames many machine learning problems as the task of comparing probability distributions, model representations, or unnormalized energies, and presents principled, scalable tools for doing so. The talk covers three threads. First, how Riemannian geometry yields a natural local comparison operator across probabilistic models, neural network latent representations, and empirical data manifolds. Second, density ratio estimation through infinitesimal classification along probability paths, introducing Conditional Time Score Matching and a vectorized variant for fast, scalable estimation. Third, learning Energy-based Models via spatiotemporal differences, an approach that unifies existing temporal and spatial algorithms while avoiding their failure modes. Together these results point to a unified perspective in which robust machine learning is achieved by constructing reasonable comparisons. The one-hour online session is suited to researchers and practitioners working in probabilistic modeling, representation learning, and generative models.