Silicon Valley Seminar October

Online

Date: Thursday, October 17
Time: 5:30 – 7:00 pm

If you would like to attend, register in advance for this webinar.

After the webinar, you are invited to talk to the speaker and mingle with other alumni in a post-webinar social via a Zoom meeting. If you would like one of our alumni or professors to share with us his/her work/interests during a future seminar, please let me know.

We look forward to seeing you.

Peter Tong

Economics-Inspired Machine Learning
By Eric Mazumdar

In a broad range of domains, including intelligent transport systems, power systems, online marketplaces, and supply chains, machine learning algorithms are rapidly becoming the main interface among stakeholders. However, their wide-spread deployment has often led to surprising failures and unwanted consequences that stem from the fact that these algorithms are often conceived in isolation and not in consideration of the large public systems in which they will be deployed. Examples of such issues, among many others, include traffic on surface streets due to over-optimized traffic routing, car crashes and congestion due to autonomous vehicles, and unfair denial of mortgages based on biased algorithms. The need for a systematic approach to dealing with these issues has only grown more urgent with the advent of a new wave of artificial intelligence technologies. Recent years have seen impressive successes in the design of so-called foundation models for language modeling (e.g., ChatGPT), catalyzed by the ease of access to internet-scale datasets, and the availability of massive amounts of compute resources. These successes in turn have sparked entire industries focused on designing AI-powered services that aim to automate decision-making from deciding who to hire to setting bail terms in criminal trials.

In this talk I will discuss how embedding ideas from economics into machine learning and AI can give us new insights into the analysis and design of machine learning algorithms for these real-world problems. 

I will begin by highlighting a line of work showing the unintuitive behaviors that arise from the interplay between learning algorithms and strategic agents like people or other algorithms. First, I will show, both in theory and practice, how learning algorithms (including those built on top of large language models) are extremely susceptible to the gaming of data by vanishingly small groups of people, even if individuals have no effect on them in isolation. I will also present recent work on how strategic interactions can break our basic intuition of always able to improve performances based on using larger models, more data, and more compute. Surprisingly, the strategic interactions can make smaller and less expressive models yield better outcomes even over systems with access to infinite data. I will conclude with some recent work on how giving AI agents features of human decision-making, like risk-aversion, can yield reproducible and efficient algorithms for game playing.

Eric is an Assistant Professor in Computing and Mathematical Sciences and Economics at Caltech. His research lies at the intersection of machine learning and economics where he is broadly interested in developing the tools and understanding necessary to confidently deploy machine learning algorithms into societal-scale systems. Eric is the recipient of a NSF Career Award and was a fellow at the Simons Institute for Theoretical Computer Science for the semester on Learning in Games. He obtained his Ph.D in EECS at UC Berkeley where he was advised by Michael Jordan and Shankar Sastry, and received his B.S. in EECS at MIT.

‍Our Alumni Volunteers‍

The following alumni work together to serve you:
Avni Gandhi, Dave Adler, Jane Frommer, Mike Klein, Xinh Huynh, and Peter Tong.