Recent Progress on Grokking and Probabilistic Federated Learning

Explore recent advancements in machine learning through this one-hour seminar presented by Thang Bui, a lecturer in Machine Learning at the Australian National University. Dive into two key topics: the grokking phenomenon and probabilistic federated learning. Discover how grokking, where neural networks achieve near-perfect accuracy on validation sets long after similar performance on training sets, extends beyond neural networks to Gaussian process classification, regression, and linear regression. Examine the hypothesis that this phenomenon is governed by the accessibility of specific regions in error and complexity landscapes. Then, delve into federated training of probabilistic models, focusing on Bayesian neural networks and Gaussian processes using partitioned variational inference. Gain insights into current techniques' limitations and potential future directions in this field.