From choices to internal states

Abstract

Complex and intelligent behavior depends not just on sensory evidence but also on internal cognitive state. Ashwood et al. use a powerful statistical method to identify hidden internal states in choice data.

Figure 1 |. The GLM-HMM framework applied to mouse and human data identifies different types of internal behavioral state.

a, Mouse task schematic. Animals use a wheel to indicate which of two stimuli they perceive as having higher contrast^,. b, States found in the mouse visual data by the GLM-HMM approach. Three states describe the animals’ performance: one high-performance state that relies heavily on sensory evidence, and two states with lower reliance on the stimulus that show left or right bias. c, Human task schematic. d, The GLM-HMM finds two states for the human data. In both, subjects rely on the stimulus; in one state there is a small leftward bias and in the other a small rightward bias.