Who Needs Kernels Anyway? - Mark Herbster

Explore a 22-minute talk by Mark Herbster from the Alan Turing Institute on predicting binary matrix entries in online learning. Delve into an algorithm that scales with matrix margin complexity, comparing its performance to the Kernel Perceptron with an optimal kernel matrix. Examine how this approach applies to biclustering and understand why the presented bound is nearly optimal, differing only by a logarithmic factor.