Search Papers | Poster Sessions | All Posters
Poster B16 in Poster Session B - Thursday, August 8, 2024, 1:30 – 3:30 pm, Johnson Ice Rink
Comparing primacy effects across different temporal scales to distinguish between theories of evidence integration
Ankani Chattoraj1 (), Xuan Wen, Ralf Haefner1; 1Nvidia Corporation, 2Vanderbilt University, 3University of Rochester
Integrating noisy sensory evidence over time is an essential feature of perception, and of perceptual decision-making. While ideal observer analyses have guided our understanding of its algorithmic and neural basis, behavior often deviates from optimality by being biased towards early evidence. Conflicting theories have been proposed to explain such a primacy effect. Algorithmic level theories include the integration of evidence only until an internal bound is reached, and a confirmation bias due to approximate hierarchical inference. Biophysical theories on Marr's implementation level include neural adaptation and attractor dynamics. Here, we tested these different hypotheses by measuring the strength of the primacy effect in a visual discrimination task across a range of time scales from total stimulus durations of 416ms to 1667ms, and for a range of evidence frame durations from 17ms to 83ms. We find the primacy effect to be invariant with respect to stimulus time, but not physical time. Furthermore, a lower influence of early evidence frames when they are very short argues against neural adaptation. Together, our results favor an algorithmic explanation as proposed by approximate hierarchical inference.
Keywords: evidence integration perceptual decision-making primacy approximate inference