Date/Time
Date(s) - 02/10/2025
1:30 pm - 2:30 pm

Speaker: Martina Neumann (TU Wien)

Location: Hamilton Hall, Room 102

Title: Consistency of augmentation graph and network approximability in contrastive learning

Abstract: Contrastive learning leverages data augmentation to develop feature representation without relying on large labeled datasets. However, despite its empirical success, the theoretical foundations of contrastive learning remain incomplete, with many essential guarantees left unaddressed, particularly the realizability assumption concerning neural approximability of an optimal spectral contrastive loss solution.

In this talk, I discuss how we overcame these limitations by analyzing pointwise and spectral consistency of the augmentation graph Laplacian.

We established that, under specific conditions for data generation and graph connectivity, as the augmented dataset size increases, the augmentation graph Laplacian converges to a weighted Laplace-Beltrami operator on the natural data manifold.

Consequently, these consistency results give way to a robust framework for establishing neural approximability, directly resolving the re