1

Disentangling Safe and Unsafe Image Corruptions via Anisotropy and Locality
State-of-the-art machine learning systems are vulnerable to small perturbations to their input, where small is defined according to a …
Ramchandran Muthukumar, Ambar Pal, Jeremias Sulam, Rene Vidal
PDF Poster Video DOI
Sparsity-aware generalization theory for deep neural networks
We show novel generalization bounds for deep neural networks that account for sparsity in intermediate layer activations of deep neural networks. Our analysis demonstrates fundamental trade-offs between sparsity and generalization.
Ramchandran Muthukumar, Jeremias Sulam
PDF Cite Poster Slides DOI
Sparsity-aware generalization theory for deep neural networks
Adversarial Robustness of Supervised Sparse Coding
Several recent results provide theoretical insights into the phenomena of adversarial examples. Existing results, however, are often …
Ramchandran Muthukumar, Jeremias Sulam
PDF Cite Code Poster Slides Video DOI