A Neuro-Fuzzy Model of Time-Varying Decision Boundaries

TL;DR

This paper introduces a fuzzy logic-based computational model for decision making that captures how humans use time-varying decision boundaries, emphasizing heuristic rules rather than optimal strategies, and fits individual data using back-propagation.

Contribution

It proposes a novel neuro-fuzzy framework for modeling decision boundaries without explicit thresholds, fitting individual behavior with a back-propagation algorithm.

Findings

Participants used time-varying boundaries from the start of the task.

Fuzzy model parameters reflect individual differences in performance.

The framework captures heuristic decision strategies effectively.

Abstract

In a recent study, we reported the results of a new decision making paradigm in which the participants were asked to balance between their speed and accuracy to maximize the total reward they achieve during the experiment. The results of computational modeling provided strong evidence suggesting that the participants used time-varying decision boundaries. Previous theoretical studies of the optimal speed-accuracy trade-off suggested that the participants may learn to use these time-varying boundaries to maximize their average reward rate. The results in our experiment, however, showed that the participants used such boundaries even at the beginning of the experiment and without any prior experience in the task. In this paper, we hypothesize that these boundaries are the results of using some heuristic rules to make decisions in the task. To formulate decision making by these heuristic rules as a computational framework, we use the fuzzy logic theory. Based on this theory, we propose a new computational framework for decision making in evidence accumulation tasks. In this framework, there is no explicit decision boundary. Instead, the subject's desire to stop accumulating evidence and responding at each moment within a trial and for a given value of the accumulated evidence, is determined by a set of fuzzy "IF-TEHN rules". We then use the back-propagation method to derive an algorithm for fitting the fuzzy model to each participant's data. We then investigate how the difference in the participants' performance in the experiment is reflected in the difference in the parameters of the fitted model