Inferring decoding strategy from choice probabilities in the presence of noise correlations

Ralf Haefner,Matthias Bethge,Jakob Macke,Sebastian Gerwinn

doi:10.1038/npre.2012.7014.1

Abstract

AbstractThe activity of cortical neurons in sensory areas covaries with perceptual decisions, a relationship often quantified by choice probabilities. While choice probabilities have been measured extensively, their interpretation has remained fraught with difficulty. Here, we derive the mathematical relationship between choice probabilities, read-out weights and noise correlations within the standard neural decision making model. Our solution allows us to prove and generalize earlier observations based on numerical simulations, and to derive novel predictions. Importantly, we show how the read-out weight profile, or decoding strategy, can be inferred from experimentally measurable quantities. Furthermore, we present a test to decide whether the decoding weights of individual neurons are optimal, even without knowing the underlying noise correlations. We confirm the practical feasibility of our approach using simulated data from a realistic population model. Our work thus provides the theoretical foundation for a growing body of experimental results on choice probabilities and correlations.

Highlights

Understanding how external stimuli give rise to sensory percepts and how individual sensory neurons support this process remain central questions of systems neuroscience
Our result shows that a recent conjecture by [24] – again based on numerical simulations – that choice probability (CP) depend on the correlation structure only through c c is true for large neuronal populations
The discovery of [3] that the response of individual sensory neurons is correlated with the animal’s behavior even when there is no signal in the stimulus has since been replicated for different decision paradigms and in different sensory cortices ( [3, 5, 11, 12, 12, 15, 20, 21, 25, 28, 30, 34, 35] with reviews in [14, 24])

Summary

Introduction

Understanding how external stimuli give rise to sensory percepts and how individual sensory neurons support this process remain central questions of systems neuroscience. One of the crucial requirements for the claim that a particular group of neurons plays a critical role in the generation of a perceptual event is that ”Fluctuations in the firing of some set of the candidate neurons to the repeated presentation of identical external stimuli should be predictive of the observer’s judgement on individual stimulus presentations” ( [29]) Such correlations between the noise fluctuations in a single neuron’s firing rate and the subject’s perceptual decision have been found in many areas (e.g. V1: [15], V2: [25], IT: [35], MT: [3,11,12,15,20,21,28,30,34], MST: [5], VIP: [11]). A neuron that itself is not total evidence

Methods

Results

Conclusion