Mengting Fang¹, Alan Stocker¹; ¹University of Pennsylvania

Perceptual decision-making often involves sequential evidence accumulation. Previous work has shown that category-level stimulus representations can play an important role in perceptual inference, even when not explicitly required. Here, we conducted a visual discrimination task to investigate how categorical representations can affect sequential evidence accumulation. Subjects discriminated the angular position (CW/CCW) of an unknown source relative to a reference based on 8 stimulus samples drawn from a Gaussian with fixed variance centered at the source position. Stimuli were presented in rapid sequence (150ms ISI). Subjects reported their categorical choice by pressing the corresponding button on a gamepad. After each trial, visual feedback displayed both the correct category and the source position. The reference was adjusted using a staircase procedure. All subjects performed the task under two conditions. In the first condition, they were asked to make a preliminary decision based on partial evidence within a 1.75s time-window, before then making their final choice after seeing all samples. The preliminary decision occurred either before the 1st sample (upfront guess) or after the 2nd, 4th, or 6th sample in the sequence. The four choice positions were randomly interleaved in each block. In the second condition (control), subjects were tested with the exact same sample sequences and reference positions as in the first condition, but simply maintained center fixation instead of making a preliminary decision. In contrast to the first condition, the reference was only shown for the final decision. Both conditions were tested in alternating blocks. Our results show that being engaged in a preliminary decision against a reference significantly improves subjects’ final decision performance compared to the control condition. This suggests that the formation of categorical stimulus representations may be crucial for accurate and robust sequential evidence accumulation.

Acknowledgements: This work was supported by the NSF CRCNS grant IIS-1912232 to A.A.S