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

Koopman-based generalization bound: New aspect for full-rank weights

Perception & Robotics Group at UMD

Jokyokai2

Stochastic Alternating Direction Method of Multipliers

A cortical information bottleneck during decision-making

Continuum Modeling and Control of Large Nonuniform Networks

How does unlabeled data improve generalization in self training

Learning group variational inference

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

Design principles for lifelong learning AI accelerators

Publications - Jiatao Gu

PAC-Bayesian Bound for Gaussian Process Regression and Multiple Kernel Additive Model

Continuum Modeling and Control of Large Nonuniform Networks