activity
20162022
most citedImproving Robustness Without Sacrificing Accuracy with Patch Gaussian Augmentation

93 citations · 159 across the 7 of their papers we have counts for

collaborators

16 papers

cs.LG202225 cited

Architecture Matters in Continual Learning

Seyed Iman Mirzadeh, Arslan Chaudhry, Dong Yin +4

A large body of research in continual learning is devoted to overcoming the catastrophic forgetting of neural networks by designing new algorithms that are robust to the distributi…

cs.LG20211 cited

An Instance-Dependent Simulation Framework for Learning with Label Noise

Keren Gu, Xander Masotto, Vandana Bachani +3

We propose a simulation framework for generating instance-dependent noisy labels via a pseudo-labeling paradigm. We show that the distribution of the synthetic noisy labels generat…

cs.LG20212 cited

Improved Regret Bound and Experience Replay in Regularized Policy Iteration

Nevena Lazic, Dong Yin, Yasin Abbasi-Yadkori +1

In this work, we study algorithms for learning in infinite-horizon undiscounted Markov decision processes (MDPs) with function approximation. We first show that the regret analysis…

cs.LG202018 cited

The Effectiveness of Memory Replay in Large Scale Continual Learning

Yogesh Balaji, Mehrdad Farajtabar, Dong Yin +2

We study continual learning in the large scale setting where tasks in the input sequence are not limited to classification, and the outputs can be of high dimension. Among multiple…

cs.LG2020

A maximum-entropy approach to off-policy evaluation in average-reward MDPs

Nevena Lazic, Dong Yin, Mehrdad Farajtabar +4

This work focuses on off-policy evaluation (OPE) with function approximation in infinite-horizon undiscounted Markov decision processes (MDPs). For MDPs that are ergodic and linear…

cs.LG2020

Optimization and Generalization of Regularization-Based Continual Learning: a Loss Approximation Viewpoint

Dong Yin, Mehrdad Farajtabar, Ang Li +2

Neural networks have achieved remarkable success in many cognitive tasks. However, when they are trained sequentially on multiple tasks without access to old data, their performanc…