Greenstreet, Francesca; (2022) Identification of action prediction error as a value free dopaminergic teaching signal in the auditory striatum. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text ThesisCorrectionsFrancescaGreenstreet.pdf - Accepted Version Download (105MB) | Preview

Abstract

To repeat actions that lead to good outcomes and to avoid repeating actions that lead to undesirable outcomes, animals must learn from experience. There is a substantial amount known regarding how the brain learns from experience to maximise rewards. This is believed to be achieved through dopamine neurons broadcasting prediction error signals about reward (RPE) to the striatum. However, dopamine neurons have been shown to not only encode RPE. Therefore an important outstanding question is how the diversity of dopamine responses can be reconciled with their hypothesised role as a teaching signal for learning. In particular, what is the role of dopamine signals that are unrelated to reward? In this thesis I firstly addressed the question of how dopamine signals can act as teaching signals for learning in different parts of the striatum. To achieve this, I recorded dopamine signals in the ventral striatum (VS) and auditory striatum (AudS) of mice learning an auditory association task. I discovered that dopamine in the VS displayed characteristics of an RPE teaching signal, consistent with the canonical theory. However, dopamine in the AudS was similar to what might be expected for a value-free teaching signal, which is consistent with recent computational models of habitual learning as being independent of any reward. I therefore tested whether dopamine signals in the VS and AudS were modulated by behavioural manipulations in a manner consistent with being a value-based and value-free teaching signal respectively. I found evidence that supported this hypothesis, which was further corroborated by simulations of value-free and value-based learning in the same tasks. These findings suggest that the AudS may be learning using a value-free learning rule, whereas the VS learns value estimates using RPE.

Download activity - last month

Export as XMLJSONCSV

Download activity - last 12 months

Export as JSONCSVXML

Downloads by country - last 12 months

Export as JSONCSVXML

Archive Staff Only

View Item