Explore benchmark design and prior-independent optimization in this 24-minute IEEE conference talk. Delve into the analysis of information-restricted algorithms, Bayesian, prior-independent, and prior-free analyses, and the characteristics of effective benchmarks. Examine normalized benchmarks, benchmark resolution, and the main theorem's implications. Investigate prior-independent mechanism design, optimal prior-independent mechanisms, and heuristic benchmark optimization. Gain insights from authors Jason Hartline, Aleck Johnsen, and Yingkai Li from Northwestern University as they present their research on this topic.
Overview
Syllabus
Intro
Analysis of Information Restricted Algorithms
Three Analyses for Information Restricted Algorithms
Outline
Bayesian, Prior-independent, and Prior-free Analyses
What makes a good benchmark? Question: What makes a good benchmark? benchmark comparison for two-server problem (Boyar, Irani, Larsen, 15)
Normalized Benchmarks
Benchmark Resolution
Discussion of Main Theorem
Prior-independent Mechanism Design
The Optimal Prior-independent Mechanisms Mechanism Design Setting
Heuristic Benchmark Optimization
Conclusions
Taught by
IEEE FOCS: Foundations of Computer Science
