Search results
My primary research interests are in learning theory, (very) broadly defined, including quantum information theory, the science of large foundation models, and high-dimensional statistics. I particularly like applications of analysis and analytic techniques to TCS problems.
Oct 27, 2024 · Jerry Li (Microsoft Research and University of Washington) Program: The workshop will be located in the Lasalle Room on the 15th floor of voco: Chicago Downton. Zoom link for remote participants
In this class, we will survey a number of recent developments in the study of robust machine learning, from both a theoretical and empirical perspective. Tentatively, we will cover a number of related topics, both theoretical and applied, including: Learning in the presence of outliers.
Sitan Chen, Jerry Li, uanzhiY Li, Anru R. Zhang Proceedings of the 55th ACM Symposium on Theory of Computing (STOC 2023) REAP: A Large-Scale Realistic Adversarial Patch Benchmark
PrincipledApproachestoRobustMachineLearning andBeyond by Jerry Zheng Li SubmittedtotheDepartmentofElectricalEngineeringandComputer Science ...
Theorem 1(small s): The expected bit length of the quantized gradient is. 32+ + log . Theorem 2(large s): For = ,there exists an encoding such that the expected bit length of the quantized gradient is 32+2.8⋅ . Original: 32n bits.
Robustness in the context of machine learning can mean a million things to a million people. In this course, we wish to try to rigorously study the role of corruption in machine learning. To do so, it will be helpful to consider the three following questions:
The SprayList: A Scalable Relaxed Priority Queue by Jerry Zheng Li B.S.,UniversityofWashington(2013) SubmittedtotheDepartmentofElectricalEngineeringandComputer
Our analysis builds on the classical results for fixed design linear regression. In linear regression, the generative model is exactly of the form described in (1), except that f is restricted to be a 1-piecewise linear function (as opposed to a k-piecewise linear function), i.e., f(x) = h ; xi for some unknown .
Jerry Li MIT jerryzli@csail.mit.edu Ludwig Schmidt MIT ludwigs@mit.edu ABSTRACT Histograms are among the most popular structures for the succinct summarization of data in a variety of database applications. In this work, we provide fast and near-optimal algorithms for approximating arbitrary one di-mensional data distributions by histograms.
