Public Presentation - Jason Harman, Ph.D. Faculty Candidate - Psychology Department
Candidate for a faculty position in the Department of Psychology
Jason L. Harman, PhD
Department of Psychology
Louisiana State University
Process Models and Simple Rules: New Models and Methods to Improve Decision Science and Machine Learning
In this talk, I review my recent work on Lexicographic Instance Based Learning (LIBL) and Multi-Criteria Model Comparison (MCMC). LIBL is a cognitive process model of human decision making that can account for all know decision making anomalies (deviations between human behavior and rational choice). The field of Judgment and Decision Making has a long tradition of discovering unique and sometimes contradictory behavioral anomalies leading to a proliferation of models designed to account for some anomalies while remining silent on others. LIBL combines dynamic memory-based choice mechanisms with heuristic encoding rules to explain choice behavior in 14 decision anomalies across multiple paradigms. Based on insights gained from LIBL’s performance in a choice prediction competition, I developed MCMC, a new model comparison procedure, to address shortcomings in scientific prediction competitions. MCMC is designed to improve insights gained from scientific prediction competitions, and model comparison in general, by comparing models across multiple scientific and theoretic criteria in a single procedure without trade-offs. The procedure has multiple advantages compared to current practices and in a recent Machine Learning (ML) competition, a simple prediction rule based on the tenants of MCMC outperformed a diverse set of ML models for personnel selection decisions that do not discriminate based on race and gender. Future applications across multiple fields will also be discussed.
Event Snapshot
When and Where
Who
Open to the Public
Interpreter Requested?
No