Research on Systems and Networks

[Home]

Algorithms for Machine Learning

We develop efficient algorithmic methods for machine learning. Examples include neural compression via coresets, federated learning with efficient communication and coresets for k-means clustering and SVD.

Relevant Publications:

OPT'20,
Direction Matters: On the Implicit Regularization Effect of Stochastic Gradient Descent with Moderate Learning Rate
with Jingfeng Wu, Difan Zou, Quanquan Gu
Full version here

IEEE BigData'21, Sketch and Scale: Geo-distributed tSNE and UMAP
with Viska Wei, Nikita Ivkin, Alexander Szalay
Full version here

ICML'20, FetchSGD: Communication-Efficient Federated Learning with Sketching
with Daniel Rothchild, Ashwinee Panda, Enayat Ullah, Nikita Ivkin, Ion Stoica, Joseph Gonzalez, Raman Arora
Full version here
Code here

ICML'20, On the Noisy Gradient Descent that Generalizes as SGD
with Jingfeng Wu, Wenqing Hu, Haoyi Xiong, Jun Huan, Zhanxing Zhu
Full version here

ICML'20, Obtaining Adjustable Regularization for Free via Iterate Averaging
with Jingfeng Wu, Lin Yang
Full version here

ICML'20, Schatten Norms in Matrix Streams: Hello Sparsity, Goodbye Dimension
with Aditya Krishnan, Roi Sinoff, Robert Krauthgamer
Full version here

ICML'20, Coresets for Clustering in Graphs of Bounded Treewidth
with Daniel Baker, Lingxiao Huang, Shaofeng H.-C. Jiang, Robert Krauthgamer, Xuan Wu
Full version here

OPT-ML'19, Obtaining Regularization for Free via Iterate Averaging
with Jingfeng Wu, Lin Yang
Full version here

RANDOM'19, Streaming Coresets for M-Estimators
with Dan Felmdan, Harry Lang, Daniela Rus
Full version here

UAI'19, Online Factorization and Partition of Complex Networks by Random Walk
with Lin F. Yang, Tuo Zhao, Mengdi Wang
Full version here

ICML'19, Coresets for Ordered Weighted Clustering
with Shaofeng Jiang, Robert Krauthgamer, Xuan Wu
Full version here

NeurIPS'19, Communication-efficient Distributed SGD with Sketching
with Nikita Ivkin, Daniel Rothchild, Enayat Ullah, Ion Stoica, Raman Arora
Full version here
Code here

NeurIPS'18, Differentially Private Robust Low-Rank Approximation
with Raman Arora, Jalaj Upadhyay
Full version here

NeurIPS'18, The Physical Systems Behind Optimization Algorithms
with Lin F. Yang, Raman Arora, Vladimir Braverman, Tuo Zhao
Full version here

ICML, Matrix Norms in Data Streams: Faster, Multi-Pass and Row-Order
with Stephen R. Chestnut, Robert Krauthgamer, Yi Li, David P. Woodruff, Lin F. Yang
Full version here

OPT, Dynamic Factorization and Partition of Complex Networks
with Lin F. Yang, Tuo Zhao, Mengdi Wang
Full version arXiv

ICML, Clustering High Dimensional Dynamic Data Streams
with Gereon Frahling, Harry Lang, Christian Sohler, Lin F. Yang
Full version here

CALDAM, Accurate Low-Space Approximation of Metric k-Median for Insertion-Only Streams
with Harry Lang, Keith Levin
Full version here

SODA, Clustering Problems on Sliding Windows
with Harry Lang, Keith Levin, Morteza Monemizadeh
Full version here