CSE Dissertation Proposal: Mithun Chakraborty

Sep 12, 2016
1 p.m.
3 p.m.
Jolley Hall, Room 309

"On The Aggregation Of Subjective Inputs From Multiple Sources"

Mithun Chakraborty
Advisor: Sanmay Das

When we have a population of individuals or artificially intelligent agents possessing diverse subjective inputs (e.g. predictions, opinions, etc.) about a common topic, how should we collect and combine them into a single judgment or estimate? This has long been a fundamental question across disciplines that concern themselves with forecasting and decision-making, and has attracted the attention of computer scientists particularly on account of the proliferation of online platforms for electronic commerce and the harnessing of collective intelligence. Looking at this problem through the lens of computational social science, I intend to divide my dissertation into three main parts: (1) Incentives in information aggregation: In this segment, I will study mechanisms for elicitation and combination of private information from strategic participants, particularly crowd sourced forecasting tools called prediction markets. I will show that (a) when a prediction market implemented with a widely used family of algorithms called market scoring rules (MSRs) interacts with myopic risk-averse traders, the price process behaves like an opinion pool, a classical family of belief combination rules, and (b) in an MSR-based game-theoretic model of prediction markets where participants can influence the predicted outcome but some of them have a non-zero probability of being non-strategic, the equilibrium is one of two types, depending on this probability -- either collusive and uninformative or partially revealing; (2) Aggregation with non-strategic agents: In this part, I will be agnostic to incentive issues, and focus on algorithms that uncover the ground truth from a sequence of noisy versions. In particular, I will present the design and analysis of an approximately Bayesian algorithm for learning a real-valued target given access only to censored Gaussian signals, that performs asymptotically almost as well as if we had uncensored signals; (3) Market making in practice: This component will deal with practical aspects of aggregation mechanisms deployed "in the wild". I will describe an adaptation of an MSR to a financial market setting called a continuous double auction and its experimental evaluation in a simulated market ecosystem.

In summary, my overarching goal is to develop a richer and deeper understanding of methods for subjective input aggregation, and identify and address challenges encountered in practical applications of these techniques.