Direct Observation of the Stochastic Drift-Diffusion Process Underlying Single Perceptual Decisions in the Primate Brain

Core Concepts

The neural signals in the lateral intraparietal area (LIP) of the primate brain directly represent the stochastic drift-diffusion process that determines the variability in choice and reaction time during perceptual decision-making.

Abstract

The study provides the first direct evidence for a drift-diffusion signal underlying single perceptual decisions in the primate brain. The authors recorded simultaneously from hundreds of neurons in the lateral intraparietal area (LIP) of monkeys while they performed a random dot motion discrimination task.
Key insights:

Using various dimensionality reduction techniques, the authors show that a scalar drift-diffusion signal can be detected in the population activity on individual trials. This signal satisfies the criteria for the decision variable that controls the choice and reaction time.
The drift-diffusion signal is dominated by a small subpopulation of neurons with response fields overlapping one of the choice targets, consistent with previous single-neuron studies.
The authors also identify direction-selective neurons in LIP that appear to represent the momentary evidence, which is then integrated by the neurons representing the decision variable.
The findings establish that the ramp-like average firing rates observed in previous studies arise from the stochastic drift-diffusion process on single trials, rather than other potential processes.
The ability to resolve the drift-diffusion signal on single trials was enabled by the use of high-density Neuropixels probes, which allowed simultaneous recording from hundreds of neurons in the primate brain.

Stats

"The choice probabilities and the distribution of reaction times (RT) are well described (Fig. 1b, traces) by a bounded drift-diffusion model (Fig. 1c)."
"The single-trial traces do not rise monotonically as a function of time but meander and tend to spread apart from each other vertically."
"The autocorrelation between an early and a later sample from the same diffusion trace is also clearly specified for unbounded diffusion."

Quotes

"We show that a single scalar quantity, derived from the weighted sum of the population activity, represents a combination of deterministic drift and stochastic diffusion."
"Moreover, we provide direct support for the hypothesis that this drift-diffusion signal approximates the quantity responsible for the variability in choice and reaction times."
"The population-derived signals rely on a small subset of neurons with response fields that overlap the choice targets. These neurons represent the integral of noisy evidence."

Key Insights Distilled From

Direct observation of the neural computations underlying a single decision

by Steinemann,N... at www.biorxiv.org 05-04-2022

https://www.biorxiv.org/content/10.1101/2022.05.02.490321v6

Deeper Inquiries

How might the representation of the decision variable in LIP change in tasks with more than two choice alternatives?

In tasks with more than two choice alternatives, the representation of the decision variable in LIP may need to accommodate the increased complexity of the decision-making process. The dimensionality of the decision variable may need to expand to encompass the additional choices, potentially requiring a higher-dimensional representation in the neural population activity. The coding directions in the neuronal state space may need to be redefined to capture the expanded set of alternatives, potentially leading to a more complex and multidimensional representation of the decision variable. Additionally, the integration of noisy evidence from direction-selective neurons in LIP may need to be adapted to account for the increased number of possible choices, potentially altering the dynamics of the decision process and the way in which the decision variable is computed.

What are the potential limitations or confounds of using dimensionality reduction techniques to identify the decision variable in neural population activity?

While dimensionality reduction techniques can be powerful tools for identifying patterns and structures in neural population activity, there are several potential limitations and confounds to consider:

Loss of Information: Dimensionality reduction techniques, such as PCA, may lead to a loss of information when reducing the high-dimensional neural activity to a lower-dimensional space. This loss of information could potentially obscure important features of the neural representation of the decision variable.

Assumptions of Linearity: Many dimensionality reduction techniques assume linearity in the data, which may not always hold true for neural population activity. Non-linear relationships and interactions between neurons may not be captured accurately by linear dimensionality reduction methods.

Curse of Dimensionality: In high-dimensional spaces, the curse of dimensionality can make it challenging to interpret the results of dimensionality reduction techniques. The increased complexity and sparsity of data in high-dimensional spaces can lead to difficulties in identifying meaningful patterns and structures.

Overfitting: Dimensionality reduction techniques may be prone to overfitting, especially when dealing with noisy or sparse data. Overfitting can result in the identification of spurious patterns or structures that do not reflect the true underlying neural representation of the decision variable.

Interpretability: While dimensionality reduction techniques can provide a compact representation of neural population activity, the reduced dimensions may be difficult to interpret in terms of the underlying neural mechanisms. Understanding the biological relevance of the reduced dimensions can be a challenge.

Could the direction-selective neurons in LIP that represent the momentary evidence play a causal role in the decision process, or are they merely providing input to the neurons representing the decision variable?

The direction-selective neurons in LIP that represent the momentary evidence could potentially play a causal role in the decision process by providing critical input to the neurons representing the decision variable. These neurons are thought to encode the direction and strength of motion, serving as the noisy momentary evidence that is integrated within LIP to form the decision variable. While they may not directly represent the decision variable itself, their activity likely influences the computation of the decision variable and ultimately the choice and reaction time on each trial.
The causal role of these direction-selective neurons in the decision process could be further elucidated through experimental manipulations, such as targeted perturbations or inactivation of these neurons. By selectively modulating the activity of these neurons, researchers could investigate their specific contributions to the decision-making process and determine whether they are essential for the accurate computation of the decision variable. Additionally, examining the temporal dynamics of these neurons in relation to the emergence of the decision variable could provide insights into their causal role in decision-making.