Neural mechanisms for decisions based on sequential samples of evidence
MetadataShow full item record
Difficult decisions often require evaluation of samples of evidence that are acquired sequentially. If samples of evidence differ in their reliability and an unlimited number of samples can be acquired sequentially, the optimal decision strategy is to accumulate evidence in units of log likelihood ratio (logLR) until the accumulated evidence exceeds a predefined threshold, or bound. For this dissertation, I hypothesized that the brain makes probabilistic inferences and decisions by such bonded accumulation of evidence. I trained monkeys to make decisions based on a sequential presentation of visual shapes. Shapes conferred probabilistic evidence that differed in predicting which target would furnish a reward. The stream of shape presentation continued until the monkeys indicated their decisions by eye movements. I found that the monkeys terminated their decisions when the cumulative logLR reached a stereotyped bound, on average. Firing rates of neurons in the lateral intraparietal area (LIP) reflected the accumulated evidence (i.e., the cumulative logLR) and also reached a stereotyped level before the monkeys committed to a decision. LIP activity explained the monkeys' choices and reaction times that reasonably approximated those expected by the optimal strategy. Psychophysical bounds for decision termination often decline in a time-dependent manner, which has been hypothesized to reflect a decision-making strategy that incorporates the cost of elapsed time. In the second part of my dissertation, I tested this hypothesis by manipulating time-cost in a perceptual choice reaction-time task. Specifically, I introduced provisional deadlines while human subjects made decisions about the direction of a random-dot motion stimulus. The pattern of choices and reaction times were explained by bounded accumulation of evidence. Termination bounds were found to decline as a function of time, which was exaggerated in response to the time-cost manipulation. In most of the previous studies of decision making in LIP, the monkeys were aware of the specific eye movement that would result from their decisions while they deliberated on the decision. Thus, it remained unknown whether evidence accumulation in LIP neurons was an abstract representation of the decision itself or a reflection of the impending eye movement. I modified the sequential sampling task, by preventing the monkey from planning a specific eye movement while it accumulated evidence. LIP neurons did not show evidence-dependent activity during the early phase of evidence accumulation period although the evidence provided in this phase affected the monkey's ultimate choice. The result suggests that LIP activity does not represent accumulated evidence per se, but only in preparation for an impending saccade. These experiments demonstrate that a bounded accumulation of evidence explains various types of decision-making such as probabilistic inference and perceptual discrimination. For monkeys, the bounded accumulation of evidence is represented in LIP activity when decisions are formed in parallel with a preparation for a specific eye movement.