Sen, Suvrajeet (University of Southern California)

Phone:
Fax: 
E-mail: s.sen@usc.edu
Website: https://sites.google.com/site/uscdatadrivendecisions/home

Topics

Big Data and Big Decisions

Over the past decade, the world of Statistical and Machine Learning have made dramatic in-roads into some of the more challenging AI problems ranging from speech recognition and natural language processing, to bio and health informatics.   Both supervised and unsupervised learning methods have exploded in daily use for applications covering business analytics, e-commerce, educational/tutoring systems, and others.   In many cases, new models and algorithms have been developed so that the results of learning are also easier to interpret (for a human decision maker).  The partnership between AI and human cognition is not new, but the widespread success in recent years has transformed the way we do business today.  The combination of modern informatics and high dimensional statistics has often been credited with this transformation.  This lecture will not only highlight some successes of Big Data, but also explore settings where human cognition may not provide the best test of decision quality.  This new class of problems involves not only Big Data, but also Big Decisions.  For instance, many states in the U.S. have committed to transforming the electricity grid to one where a large fraction of electricity will be produced by renewable resources.  In California, the goal is to have 50% of the electric power generated using renewable resources by 2030.  As of now, it is unclear whether this is feasible without a great deal of over-generation.  Of course, the latter goes against the goal of reducing emissions.   Since these trade-offs involves many “moving parts” (multiple markets, technologies and constraints) the human brain is not the ideal computational device for separating good decisions from bad ones.  Such cases arise in many other socio-technical systems, such as transportation, water resources, and others.  This lecture will explore the continuum between Big Data and Big Decisions.  We will also discuss how Big Data can be integrated into an OR/MS Curriculum. (Elementary)

Learning Enabled Optimization

At the core of Statistical/Machine Learning are large scale optimization algorithms which help parameter choices in statistical models. This aspect of learning may be looked upon as "Optimization Enabled Learning". This lecture is presents an alternative viewpoint in which learning methods inform optimization models.  Statistics has a long history of using mathematical models to separate systematic trends from noise. We propose using these statistical ideas for modeling decision problems in which prediction of future trends and errors can be integrated within optimization models.  This new class of models finds applications in Analytics, especially in studying the Analytics of Things (AoT), and Integrating Predictive Modeling with Prescriptive Analytics.  We discuss a variety of open problems for Learning Enabled Optimization. (Intermediate)

Algorithmic Approaches for Learning Enabled Optimization

At the core of Statistical/Machine Learning are large scale optimization algorithms which help parameter choices in statistical models. This aspect of learning may be looked upon as "Optimization Enabled Learning". This lecture is presents an alternative viewpoint in which learning methods inform optimization models.  Statistics has a long history of using mathematical models to separate systematic trends from noise. We propose using these statistical ideas for modeling decision problems in which prediction of future trends and errors can be integrated within optimization models.  We present new concepts of optimality such as statistical optimality, as well as metrics which help choose decisions which avoid "over-fitting" and improve generalizability of optimization models. (Advanced)

Background: