3 papers
cs.LG2026
Beyond Linear and Overcomplete Regimes: A Mean-Field Analysis of Bottleneck Autoencoders
Santanu Das, Ramyak Bilas, Pascal Esser +1
Autoencoders (AEs) learn low-dimensional representations by mapping data into a latent space while minimizing reconstruction error. Despite their empirical success, theoretical und…
cs.LG2026
Tractable Gaussian Phase Retrieval with Heavy Tails and Adversarial Corruption with Near-Linear Sample Complexity
Santanu Das, Jatin Batra
Phase retrieval is the classical problem of recovering a signal from its noisy phaseless measurements (where …
cs.LG2025
A direct proof of a unified law of robustness for Bregman divergence losses
Santanu Das, Jatin Batra, Piyush Srivastava
In contemporary deep learning practice, models are often trained to near zero loss i.e. to nearly interpolate the training data. However, the number of parameters in the model is u…