Iclr2020: Compression based bound for non-compressed network

Iclr2020: Compression based bound for non-compressed network: unified generalization error analysis of large compressible deep neural network - Download as a PDF or view online for free
1) The document presents a new compression-based bound for analyzing the generalization error of large deep neural networks, even when the networks are not explicitly compressed. 2) It shows that if a trained network's weights and covariance matrices exhibit low-rank properties, then the network has a small intrinsic dimensionality and can be efficiently compressed. 3) This allows deriving a tighter generalization bound than existing approaches, providing insight into why overparameterized networks generalize well despite having more parameters than training examples.

ICLR 2020

Publications - OATML

NeurIPS2020 (spotlight)] Generalization bound of globally optimal non convex neural network training: Transportation map estimation by infinite dimensional Langevin dynamics

How does unlabeled data improve generalization in self training

On the Resilience of Deep Learning for reduced-voltage FPGAs

Higher Order Fused Regularization for Supervised Learning with Grouped Parameters

PDF) FEDZIP: A Compression Framework for Communication-Efficient Federated Learning

Minimax optimal alternating minimization \ for kernel nonparametric tensor learning

Publications - Jiatao Gu

Iclr2020: Compression based bound for non-compressed network: unified generalization error analysis of large compressible deep neural network