Data Science Through a Geometric Lens


Time Speaker Title
9:00 Cyrus Rashtchian (UCSD) Opening Remarks
9:10 David Woodruff (CMU) Strong Coresets for k-Median and Subspace Approximation, Goodbye Dimension
9:40 Suriya Gunasekar (TTIC) Characterizing Implicit Bias in Terms of Optimization Geometry
10:10 Rebecca Willett (UChicago) Low Algebraic Dimension Matrix Completion
10:40 Hanyu Zhang (UW) From Non-parametric to Parametric: Manifold Coordinates with Physical Meaning
11:10 Coffee Break
11:30 Samory Kpotufe (Columbia) Kernel Sketching yields Kernel JL
12:00 Nina Balcan (CMU) Data Driven Clustering
12:20 Organizers Closing Remarks
12:30 Lunch

About the Workshop

Machine learning and data science applications heavily rely on geometric ideas and algorithms, such as clustering, metric embeddings, sketching, and nearest neighbor search. Unfortunately, classical computational models and worst-case analysis often fail to capture both the nuances of modern datasets and the overwhelming success of many methods in practice. Therefore, the time is ripe to focus on geometric algorithms with an eye towards relevant use cases. This workshop will introduce participants to both (i) the theoretical achievements in the area and (ii) the current and future applications of these ideas. Several excellent speakers will present their research, which should be of interest to anyone working in the union of theoretical computer science, machine learning, geometry, and statistics.


Sunday, June 23, 2019
9:00am - 12:30pm

STOC 2019
Room: West 212C