Title: Low-Budget Active Learning Through Entropic Optimal Transport

Abstract:
We address low-budget active learning, which consists in selecting a limited number of points, the coreset, such that a model can be trained to high accuracy on the selection only.
The learning problem formalizes as a combinatorial optimization problem over the hypersimplex, which is provably challenging for existing integer optimization methods and convex relaxations, for complexity and geometry reasons.
In this paper, we leverage entropic optimal transport, using the Sinkhorn divergence as the coreset selection criterion, which first allows us to get dimension-free sample complexity results, and second admits computationally efficient gradient evaluations.This opens the way to using gradient-based algorithms to rapidly compute solution candidates, with guarantees on the solution quality.
Experiments on standard image and EEG benchmarks show that our method outperforms state-of-the-art heuristics in low-budget settings.