Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster D in Poster Session D: Wednesday, August 5, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Determinants of relative value coding in artificial and biological networks

Kenway Louie¹, Shreya Sinha², Priyamvada Modak², Dino Levy³; ¹NYU Langone Health, ²New York University, ³Tel Aviv University

Presenter: Kenway Louie

Value is a fundamental variable in decision-making, integrating potential cost and benefit information into a unitary quantity to guide the selection process. While rational choice theory assumes that options are assigned absolute values, independent of other available alternatives, biological decision processes often rely on comparative evaluation and decision-related brain circuits employ a relative rather than absolute value code. While relative value coding is widely observed, it is unknown why such coding arises and what function it serves. Here, we examine what environmental and internal factors drive relative reward coding in deep neural networks trained in economic decision tasks. We find that relative value coding: (1) arises naturally in deep networks, (2) decreases when coding capacity (layer size, network depth) is expanded, and (3) increases when inputs exhibit more statistical structure (correlated variability). Furthermore, the degree of relative value coding is modulated by both internal and external noise. Together, these findings suggest that relative valuation reflects an efficient coding process, optimizing network performance under information processing constraints.

Topic Area: Decision-Making, Cognitive Control & Event Cognition