Robust Aggregation of Noisy Estimates
Abstract
Statement of Work:Algorithms for the aggregation of noisy estimates will be designed, analytical bounds on their performance will bederived, and empirical evidence for their performance in the real world will be provided.Objective:To develop new methods for aggregating noisy votes. Specifically, the input to these algorithms consists of objective estimates of the quality of alternatives, represented as noisy rankings. The noise is assumed to be randomly generated according to a noise model. The goal is to output a ranking of the alternatives that reflects their true quality as accurately as possible.Approach:The project will build on the idea of minimax estimators. That is, each algorithm will minimize the maximum expected error in estimating the ground truth, where the maximum is taken over all noise models in a given collection. Analytical bounds on error will be provided, as well as empirical validation of the performance of proposed methods on real data.Overall Merit and ONR Mission/Relevance:Effective group decision is a cornerstone to the success of any organization, including the Navy. Potential applications include the selection of a mission plan that is most likely to be successful, and the selection of a new technology to devote resources to, based on expert opinions.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 23, 2016
- Source ID
- N000141613075
Entities
People
- Ariel Procaccia
Organizations
- Massachusetts Institute of Technology
- Office of Naval Research
- United States Navy