Sources of noise during accumulation of evidence in unrestrained and voluntarily head-restrained rats

Benjamin B. Scott, Christine M. Constantinople, Jeffrey C. Erlich, David W. Tank, Carlos D. BrodyeLife 2015 · DOI

TL;DR

This paper introduces a novel visual evidence accumulation task for rats that allows for voluntary head restraint, paving the way for future neural imaging studies. They discovered that the primary source of noise in decision-making scales linearly with the number of evidence pulses, a finding that challenges the widely used drift-diffusion models and aligns more with models of numerical cognition.

What Problem Does This Paper Address?

Decision-making, even in simple tasks, is notoriously variable. While "noise" in the evidence accumulation process is a leading suspect for this variability, its precise nature and origin remain unclear. Existing models, particularly the popular drift-diffusion models (DDMs), often assume noise that scales with the square root of the evidence. However, these models struggle to explain errors on easy trials and may not fully capture the complexities of how animals integrate information. This paper aims to dissect the sources of noise in evidence accumulation by developing a new task and applying advanced modeling techniques.

Methods

The researchers developed a visual "accumulation of evidence" task where rats had to discriminate between the number of flashes presented to their left and right visual hemifields. A key innovation was adapting this task for voluntary head restraint, which is crucial for enabling cellular-resolution imaging and manipulation in future studies. They employed a signal-detection theory (SDT) framework to model the accumulation process, focusing on the relationship between the number of evidence pulses (flashes) and the noise in the internal estimate of those pulses. They also compared various accumulator models (single vs. dual, static vs. sequential sampling) and noise scaling assumptions (scalar variability vs. linear variance).

Datasets

The study utilized a large dataset of behavioral trials collected from seven rats performing the visual accumulation of evidence task. This dataset comprised over 525,000 trials. They also re-analyzed data from a previously published auditory accumulation of evidence task for comparison.

Key Insights and Ideas

  • Scalar Variability is King (for noise): The most significant finding is that the noise in the rats' internal estimates of flash numbers scaled linearly with the number of flashes, a phenomenon known as scalar variability. This is a departure from the square-root scaling predicted by standard drift-diffusion models.
  • Challenging DDM: This linear scaling of noise suggests that the noise is not independent for each individual flash, as often assumed in DDMs. This implies that current DDM frameworks might need significant revision to accurately capture decision-making processes.
  • Dual Accumulators: The data strongly favored models with two separate accumulators (one for left, one for right evidence) over a single accumulator model. This suggests that rats maintain distinct representations of evidence from each side before making a decision.
  • Voluntary Head Restraint Works: The study successfully demonstrated that rats can perform this complex cognitive task with comparable accuracy and performance characteristics while voluntarily head-restrained, validating this approach for future neuroscientific investigations.
  • Trial History Matters (but adds noise): While rats integrate evidence over the entire trial, their decisions are also subtly biased by past rewards and errors, up to three trials prior. This trial history, while potentially adaptive in other contexts, acts as an additional source of noise in this specific task.

Key Results

  • Rats performed the visual accumulation task with high accuracy (70-91% correct).
  • The standard deviation of the internal estimate of flash number scaled linearly with the number of flashes presented (scalar variability).
  • Models with two accumulators and scalar variability noise provided the best fit to the behavioral data.
  • Head-restrained rats performed the task comparably to unrestrained rats, with similar noise characteristics and accumulation time constants.
  • Trial history (reward/error) introduced a modest bias that persisted for up to three trials.

Limitations and Future Work

The authors acknowledge that their exploration of noise models was not exhaustive, particularly regarding the gradual addition of noise during stimulus presentation. They suggest that future work should focus on incorporating scalar variability into decision-making models and exploring the neural mechanisms underlying this type of noise. The development of the voluntary head-restraint task is a significant step towards enabling such investigations using advanced neuroimaging techniques. They also propose developing a reaction-time version of the task to further probe accumulation dynamics.

Summary updated

Abstract

Decision-making behavior is often characterized by substantial variability, but its source remains unclear. We developed a visual accumulation of evidence task designed to quantify sources of noise and to be performed during voluntary head restraint, enabling cellular resolution imaging in future studies. Rats accumulated discrete numbers of flashes presented to the left and right visual hemifields and indicated the side that had the greater number of flashes. Using a signal-detection theory-based model, we found that the standard deviation in their internal estimate of flash number scaled linearly with the number of flashes. This indicates a major source of noise that, surprisingly, is not consistent with the widely used 'drift-diffusion modeling' (DDM) approach but is instead closely related to proposed models of numerical cognition and counting. We speculate that this form of noise could be important in accumulation of evidence tasks generally.
BibTeX
@inproceedings{scott2015sources,
  author = {Benjamin B. Scott and Christine M. Constantinople and Jeffrey C. Erlich and David W. Tank and Carlos D. Brody},
  title = {{Sources of noise during accumulation of evidence in unrestrained and voluntarily head-restrained rats}},
  booktitle = {eLife},
  year = {2015},
  doi = {10.7554/elife.11308},
}
Figure