return to TMARC
| Publication Abstract Display | | Type: Poster | | Title: Varieties of reinforcement learning styles in HIV infection and methamphetamine dependence: A computational modeling study. | | Authors: Brown GG, Bischoff Grethe A, Jordan SJ, Tapert S, Heaton R, Cherner M, Grant I | | Date: 02-21-2019 | | Abstract:Objective: Most researchers using reinforcement learning (RL) to study dopaminergic function assume a homogeneous learning style among subjects. In this study, we use computational models of RL to test this assumption among individuals with/without HIV infection and with/without methamphetamine dependence.Participants and Methods: 60 individuals performed 2 sets of an RL task intermixing reward and punishment trials as part of a fMRI study. Six RL models represented different learning styles from random learning to complex learning involving parameters for reward and punishment learning, exploration/exploitation trade-offs, and response repetition bias. Evidence ratios compared model fits.Results: For Set 1, group was related to random vs. systematic learning, χ2(3) = 8.161, p=.045, φ = .37. 82% of random learners were HIV positive, associating HIV+ status with random learning, χ2(1) = 6.89, p=.013, φ = -.34. Among systematic learners on Set 1, models including separate reward and punishment parameters fit best. Eleven of 18 Set 1 HIV+ random learners showed evidence of learning on Set 2, contributing to a more equal distribution of random/systematic learners across the HIV positive/negative groups, χ2(1) = 0.057, p=.812, φ = -.031. Among participants with evidence of learning on Set 2, models including the repetition bias parameter best fit the sample. Of the 11 HIV+ participants who switched from Set 1 random learning to Set 2 systematic learning, 8 showed evidence of shifts to models including the bias parameter and only 2 showed the typical Set 1 learning style on Set 2. In no analysis did the inferred learning style of methamphetamine dependent individuals significantly differ from healthy controls.Conclusions: RL style varied across participants, with the principal style changing with practice. Half of the HIV infected individuals showed evidence of delayed or impaired learning that might reflect altered dopaminergic function. |
return to publications listing
|