,
11 tweets,
4 min read
Read on Twitter
Excited to share my new paper with Todd Hagen and Bob Wilson @NRDlab on suboptimalities in evidence accumulation and how they relate to pupil, now out in @NatureHumBehav rdcu.be/bqpDM
For simple perceptual decisions, previous work suggests that humans and animals can integrate evidence over time, but not optimally.
This suboptimality could arise from sources including: neuronal noise, weighting evidence unequally over time - 'integration kernel', previous trials effects, and an overall bias.
We use an auditory evidence accumulation task in 100+ human participants, in which participants listen to twenty clicks coming in from the left and the right ears over one second, and decide which side has more clicks. 
We show that people exhibit all four suboptimalities: overall noise, a 'bump' shaped integration kernel (people weigh the middle of the stimulus significantly more than the beginning or the end), ...
... previous trials effects (people tend to pick whichever side that was correct on the previous trial (reinforcement learning effect), and whichever side they did not pick on the previous trial (choice kernel)), and an overall side bias.
Some of these suboptimalities covary across the population - the higher the signal-to-noise ratio of a participant, the smaller the magnitude of previous trial effects (reinforcement learning and choice kernel).
Our pupillometry results show that only noise and integration kernel are related to the change in pupil response. Moreover, these two different suboptimalities are related to different aspects of the pupil signal.
The individual differences in pupil change are associated with individual differences in integration kernel - participants with larger change in pupil dilation during the stimulus tend to have a more uneven integration kernel.
On the other hand, the trial-by-trial fluctuations in pupil change are associated with trial-by-trial fluctuations in noise. These results suggest that different suboptimalities relate to distinct pupil-linked processes.
Finally, we fit Drift Diffusion Model (Brunton, et al. 2013) to our data using code in Yartsev, et al. 2018. We find that decision bound marginally correlates with pupil change, potentially suggesting that pupil affects the integration kernel shape by changing the bound height.
