Noise in neural value signals links preference variability, choice stochasticity and confidence

Poster No:

773 

Submission Type:

Abstract Submission 

Authors:

Raphael Le Bouc1, Gilles de Hollander1, Marcus Grueschow1, Rafael Polania2, Christian Ruff1

Institutions:

1UZH, Zürich, Switzerland, 2ETH, Zürich, Switzerland

First Author:

Raphael Le Bouc  
UZH
Zürich, Switzerland

Co-Author(s):

Gilles de Hollander  
UZH
Zürich, Switzerland
Marcus Grueschow  
UZH
Zürich, Switzerland
Rafael Polania  
ETH
Zürich, Switzerland
Christian Ruff  
UZH
Zürich, Switzerland

Introduction:

The capacity to evaluate and choose the most rewarding options in the environment is crucial for survival. However, humans show surprising variability in their preferences, confidence and choices regarding rewards, a phenomenon that lacks a comprehensive mechanistic explanation. This variability is classically attributed to the presence of random noise in behavioural measures or in the neural processes underlying behaviour. However, the specific neural computations affected by this noise remain uncertain.

Methods:

Here we provide evidence that variability in value-based decision-making arises in part from the idiosyncratic precision with which the value of each reward option is encoded in the brain. We demonstrate this using behavioural tasks that measure the individual variability in the subjective value participants assign to food items, their confidence in the expressed value, and their stochasticity in two-option choices. We then relate these measures to the precision of probabilistic neural value representations decoded using functional magnetic resonance imaging and a population receptive field model combined with a Bayesian decoder.

Results:

Reward items with more precise value representations in ventromedial prefrontal cortex were evaluated more reliably, with higher confidence, and were chosen more consistently.

Conclusions:

Thus, our findings offer a unified explanation for the imprecision observed in decision-making behaviours, which reflects not only random neural noise but a more fundamental property, the precision with which the brain encodes reward value.

Emotion, Motivation and Social Neuroscience:

Reward and Punishment 1

Modeling and Analysis Methods:

Activation (eg. BOLD task-fMRI)
Bayesian Modeling
Classification and Predictive Modeling 2

Keywords:

Cognition
Computational Neuroscience
FUNCTIONAL MRI
Other - decision-making

1|2Indicates the priority used for review

Provide references using author date format

NA