By Camilla Ucheoma Enwereuzor
As part of my MSc internship in the lab, I have spent the last couple of months diving into a paper by Ashwood et al. (2022), who investigated how observers switch between different strategies for perceptual decision-making over the course of long testing sessions. According to previous accounts (e.g., Wichmann & Hill, 2001), subjects maintain one main strategy during cognitive tasks, and any lapses (i.e., errors despite strong sensory evidence) arise independently of one another and of the time course of the experimental session. However, Ashwood et al. suggest that this view is not correct. Using a modelling approach based on Hidden Markov Models (HMMs), the authors found that mice switch between multiple strategies, or hidden “states”, during perceptual decision-making sessions (Figure 1). Importantly, the Markovian component of this approach implies that states are not independent of one another, but rather depend only on the state from the preceding trial, and can persist for many trials in a row. Let us look at how this modelling approach works in more detail.
Figure 1. Reprinted from Figure 1a-b of Histed and O’Rawe (2022). a. Schematic representation of the task for the IBL et al. (2021) mouse data. Mice turned a wheel to indicate whether a sinusoidal grating appeared on the left or right side of the screen. b. Recovered states according to a 3-state GLM-HMM: mice switched between an engaged state where they relied heavily on sensory evidence, to less engaged states that showed left or right biases.
CoCoSys lab 