Online Seminar on Mathematical Foundations of Data Science
A weekly online seminar on random topics on mathematical foundations of machine learning, statistics and optimization Sponsored by Two Sigma
Two Sigma Fellowship
Our sponsor, Two Sigma, is soliciting applications for TWO SIGMA PHD FELLOWSHIP and TWO SIGMA DIVERSITY PHD FELLOWSHIP.
To ask the speaker a question...
To ask the speaker a question, please use the Q&A function in Zoom to type your questions. The moderator (Zhuoran Yang) will collect questions through the talk and ask the questions after the talk.
Next Speaker: Itai Ashlagi, Stanford University
Title: Price Discovery in Waiting Lists: A Connection to Stochastic Gradient Descent
Abstract: Waiting lists allocate items by offering agents a choice among items with associated waiting times. These waiting times serve as prices that are determined endogenously and adjust according to the stochastic arrivals and departures of agents. We study the allocative efficiency under such dynamically adjusting prices by drawing a connection between this price adjustment process and the stochastic gradient descent optimization algorithm. We show that the loss due to price fluctuations is bounded by the granularity of price changes. Additional conditions allow us to identify markets where the loss is close to the bound or exponentially small. Our results show that a simple price adjustment heuristic can perform well, but may be slow to adjust to changes in demand because of a trade-off between the speed of adaptation and loss from price fluctuations.
Joint work with Jacob Leshno, Amin Saberi and Pengyu Qian
Bio: Itai Ashlagi is an Associate Professor at the Management Science & Engineering Department. He is interested in game theory and the design and analysis of marketplaces. He is especially interested in marketplaces, in which matching is an essential activity. markets, for which he developed mechanisms using tools from operations/cs and economics. His work influenced the practice of Kidney exchange, for which he has become a Franz Edelman Laureate. Ashlagi received his PhD in operations research from the Technion-Israel Institute of Technology. Before coming to Stanford he was an assistant professor of Operations Management at Sloan, MIT and prior to that a postdoctoral researcher at HBS. He is the recipient of the outstanding paper award in the ACM conference of Electronic Commerce 2009. His research is supported by the NSF including an NSF-CAREER award.
Upcoming Speakers
Organizers
If you have any questions, please feel free to contact the organizers.
Ethan X. Fang, Niao He, Junwei Lu, Zhaoran Wang, Zhuoran Yang, Tuo Zhao
Penn State University
ETH Zurich
Harvard University
Northwestern University
Princeton University
Georgia Institute of Technology
Previous Talks (Including Recordings)
Princeton University
Title: Statistical Inference on Membership Profiles in Large Network
Date/Time: May 19th 3pm EDT
Princeton University
Title: On the Statistical Complexity of Reinforcement Learning
Date/Time: May 26th 3pm EDT
Stanford University
Title: Distributionally Robust Optimization, Online Linear Programming and Markets for Public-Good Allocations
Date/Time: June 2nd 3pm EDT
Sloan School of Management, MIT
Title: From Stochastic Frank-Wolfe to the Ellipsoid Method: Recent Progress on Practical Optimization in Data Science and Theoretical Optimization
Date/Time: June 9th 3pm EDT
Sham Kakade
University of Washington, Seattle
Title: Representation, Modeling, and Gradient Based Optimization in Reinforcement Learning
Date/Time: June 16th 3pm EDT
University of Wisconsin-Madison
Title: Nonparametric Active Learning with Kernels and Neural Networks
Date/Time: June 23rd 3pm EDT
Columbia University
Title: Some Recent Insights on Transfer-Learning
Date/Time: July 21th 3pm EDT
UC Berkeley
Title: Optimization with Momentum: Dynamical, Variational, and Symplectic Perspectives
Date/Time: August 11th 3pm EDT
UIUC
Title: Policy Optimization for Linear Optimal Control with Guarantees of Robustness
Date/Time: August 25th 3pm EDT
Harvard University
Title: On Nearly Assumption-Free Tests of Nominal Confidence Interval Coverage for Causal Parameters Estimated by Machine Learning
Date/Time: September 1st 3pm EDT
Microsoft Research
Title: Statistical Optimization Methods for Machine Learning
Date/Time: Tuesday Sep 15th 3pm EDT
Cornell University
Title: Queueing Network Controls via Deep Reinforcement Learning
Date/Time: Friday Sep 25th 11am EDT
Princeton University
Title: Learning Dynamical Systems with Side Information
Date/Time: Friday Oct 2nd 11am EDT
Harvard University
Title: Off-Policy Estimation of Long-Term Average Outcomes with Applications to Mobile Health
Date/Time: Friday Oct 9th 11am EDT
University of Florida
Date/Time: Friday Oct 16th 12pm EDT
MIT
Title: Statistical Learning in Operations: The Interplay between Online and Offline learning.
Date/Time: Friday Nov 6th 11am EST
Columbia University
Title: Complexity of High Dimensional Sparse Functions
Date/Time: Friday Nov 13th 11am EST
Yale University
Title: Predicting Disease Risk from Genomics Data
Date/Time: Tuesday Nov 17th 11am EST
Carnegie Mellon University
Title: Preconditioning Helps: Faster Convergence in Statistical and Reinforcement Learning
Date/Time: Friday Jan 15th 11am EDT
New York University
Title: Robust Online Learning and its Applications to Assortment Optimization
Date/Time: Friday Jan 22th 4pm EDT
California Institute of Technology
Title: Competitive Control via Online Optimization
Date/Time: Friday Jan 29th 11am EDT
Massachusetts Institute of Technology
Title: Machine Learning under a Modern Optimization Lens
Date/Time: Friday Feb 5th 11am EST
University of Alberta/DeepMind
Title: Hardness of MDP Planning with Linear Function Approximation
Date/Time: Friday Feb 12th 11am EST
University of Southern California
Title: Deciphering Neural Networks through the Lens of Feature Interactions
Date/Time: Friday Feb 26th 3pm EST
Georgia Tech
Title: Stochastic Optimization Algorithms for Reinforcement Learning
Date/Time: Friday Mar 5th 11am EST
EPFL
Title: General Framework for Optimal Data-Driven Optimization
Date/Time: Friday Mar 19th 11am EST
Harvard University
Title: Real-Time Distributed Decision Making in Networked Systems
Date/Time: Friday Mar 26th 11am EDT
Microsoft Research
Title: A law of robustness for two-layers neural networks
Date/Time: Friday April 2nd 11am EDT
Massachusetts Institute of Technology
Title: Causal Inference for Panel Data with General Treatment Patterns
Date/Time: Friday Apr 16th 11am EDT
University of Texas, Austin
Title: MLE and EM Algorithms for Log-Concave Mixtures
Date/Time: Friday Apr 23th 11am EDT
Northwestern University
Title: Derivative-Free Optimization of Noisy Functions
Date/Time: Friday April 30th 11am EDT
Yale University
Title: Balancing Covaraites in Randomized Experiments
Date/Time: Friday May 7th 11am EDT
University of Chicago
Title: Machine Learning and Inverse Problems in Imaging
Date/Time: Friday May 14th 11am EDT
University of Wisconsin-Madison
Title: The role of Complexity Bounds in Optimization
Date/Time: Friday June 4th 11am EDT
Northwestern University
Title: Distributionally Robust Two- and Multi-Stage Stochastic Programming
Date/Time: Friday June 11th 11am EDT
University of North Carolina
Date/Time: Friday June 18th 11am EDT
Stanford University
Title: Simple Agent, Complex Environment: Efficient Reinforcement Learning with Agent States
Date/Time: Friday June 25th 11am EDT
Yale University
Title: Gradient Descent, Inversion, and Repair in Overparameterized Models
Date/Time: Friday July 2nd 11am EDT
Georgia Institute of Technology
Title: Statistical Inference for Spatio-Temporal Point Process
Date/Time: Friday July 16th 11am EDT
University of Chicago
Title: Optimization and Estimation in High Dimensions
Date/Time: Friday July 23rd 11am EDT
Duke University
Title: Continuous Time Contract Design and Stochastic Optimal Control
Date/Time: Friday August 6th 11am EDT
Eli Lilly
Title: Our Recent Development on Cost Constraint Machine Learning Models
Date/Time: Friday September 24th 12pm EDT
Yale University
Title: Recent Results in Planted Assignment Problems
Date/Time: Friday October 1st 11am EDT
Yale University
Title: Estimating Heterogeneous Treatment Effects Using Machine Learning
Date/Time: Friday October 22nd 11am EDT
