CCN meeting | Stefano Palminteri (École normale supérieure, France)

When
17-03-2022 from 15:00 to 16:00
Where
Henri Dunantlaan 2, room 2.2 & https://ugent-be.zoom.us/j/94014901971?pwd=QzNHR25tTFhuaXVabWY2dzRBUTE3UT09
Language
English

Contrasting range adaptation and divisive-normalization in human reinforcement learning

Contrasting range adaptation and divisive-normalization in human reinforcement learning

Context-dependent learning has been shown to lead to irrational choices in humans. This is specifically true when the options are extrapolated from their original learning context. In a previous study, we showed that this process was well captured by a dynamical range normalization model, inspired by the range-frequency theory and electrophysiological findings in monkeys. However, the two-armed bandit previously used is ill-suited to precisely characterize the functional form of context-dependence as range normalization or divisive normalization. To fill this gap, we designed a new online-based learning task simultaneously manipulating the number of options per context (2-armed bandit versus 3-armed bandit) and the range magnitude of the options, by varying their expected values. We also included an explicit valuation phase where participants had to report their estimation of each option. Behavioural and computational analyses seriously challenge divisive normalization but suggest that simple range normalization cannot account for all behavioural patterns. Together, these results shed new light on the mechanisms of context-dependent learning in humans.