H
Haoyu Zhao
Researcher at Tsinghua University
Publications - 17
Citations - 98
Haoyu Zhao is an academic researcher from Tsinghua University. The author has contributed to research in topics: Computer science & Type (model theory). The author has an hindex of 4, co-authored 12 publications receiving 50 citations. Previous affiliations of Haoyu Zhao include Princeton University.
Papers
More filters
Posted Content
Mildly Overparametrized Neural Nets can Memorize Training Data Efficiently
Rong Ge,Runzhe Wang,Haoyu Zhao +2 more
TL;DR: It is shown that neural networks can be trained to memorize training data perfectly in a mildly overparametrized regime, where the number of parameters is just a constant factor more than thenumber of training samples, and theNumber of neurons is much smaller.
Proceedings Article
BEER: Fast O(1/T) Rate for Decentralized Nonconvex Optimization with Communication Compression
TL;DR: This paper proposes BEER, which adopts communication compression with gradient tracking, and shows it converges at a faster rate of O (1 /T ) than the state-of-the-art rate, by matching the rate without compression even under arbitrary data heterogeneity.
Posted Content
Combinatorial Semi-Bandit in the Non-Stationary Environment.
TL;DR: A parameter-free algorithm is designed that achieves nearly optimal regret both in the switching case and in the dynamic case without knowing the parameters in advance.
Proceedings ArticleDOI
SoteriaFL: A Unified Framework for Private Federated Learning with Communication Compression
TL;DR: A framework called SoteriaFL is proposed, which accommodates a general family of local gradient estimators including popular stochastic variance-reduced gradient methods and the state-of-the-art shifted compression scheme, and is shown to achieve better communication complexity without sacrificing privacy nor utility than other private federated learning algorithms without communication compression.
Proceedings Article
Combinatorial Pure Exploration for Dueling Bandit
TL;DR: This paper designs a fully polynomial time approximation scheme (FPTAS) for the offline problem of finding the Condorcet winner with known winning probabilities, and uses the FPTAS as an oracle to design a novel pure exploration algorithm CAR-Cond with sample complexity analysis.