Theses Doctoral

Inferring Decision Rules from Evidence, Choice, and Reaction Times

Kang, Yul Hyoung Ryul

When a decision is made based on noisy evidence, it is often a good strategy to take multiple samples of evidence up to a threshold before committing to a choice. Such process, termed bounded evidence accumulation, have successfully explained human and nonhuman behavior (speed and accuracy of choices) and neural recordings quantitatively. In this thesis, we exploit the quantitative relationship between evidence, choice, and reaction times (inverse of speed), to infer decision rules that are not reported directly.
In Part I, we consider decisions based on one stream of evidence. In Chapter 2, we start by examining decisions that are not reported immediately but felt to be made at some point. We show that, in a perceptual decision-making task, we can predict the proportion of choices from the reported timing of covert decisions. We suggest that the awareness of having decided corresponds to the threshold-crossing of the accumulated evidence, rather than a post hoc inference or arbitrary report. For the type of decisions reported in Chapter 2 and many others, it has been suggested that the terminating threshold is not constant but decreases over time. In Chapter 3, we propose a method that estimates the threshold without any assumption on its shape.
As a step toward more complex decisions, in Part II we consider decisions based on two streams of evidence. In Chapter 4, we summarize the results from human psychophysics experiments involving simultaneous motion-color judgments. The results suggest that information bearing on two dimensions of a decision can be acquired in parallel, whereas incorporation of information into a combined decision involves serial access to these parallel streams. Here, one natural question is how complete the seriality is. In Chapter 5, we propose a method to estimate the degree of seriality of two evidence accumulation processes. Another question is whether the two streams are acquired in parallel even when the stimulus viewing duration is not limited, and hence there is no apparent advantage to parallel acquisition given the serial evidence accumulation stage. In Chapter 6, we propose a method to estimate the probability of simultaneous acquisition of two evidence streams given the choice and evidence streams.
Collectively, the work in this thesis presents new ways to study decision rules quantitatively given noninvasive measures such as the contents of the evidence stream(s), decision times, and the choice.


  • thumnail for Kang_columbia_0054D_14316.pdf Kang_columbia_0054D_14316.pdf application/pdf 5.42 MB Download File

More About This Work

Academic Units
Neurobiology and Behavior
Thesis Advisors
Shadlen, Michael N.
Ph.D., Columbia University
Published Here
November 6, 2017