Date of Award

2015

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Graduate Group

Neuroscience

First Advisor

Nicole C. Rust

Abstract

The task of finding specific objects and switching between targets is ubiquitous in everyday life. Searching for a particular object requires our brains to activate and maintain a representation of the target (working memory), identify each encountered object (object recognition), and determine whether the currently viewed object matches the sought target (decision making). The comparison of working memory and visual information is thought to happen via feedback of target information from higher-order brain areas to the ventral visual pathway. However, what is exactly represented by these areas and how do they implement this comparison still remains unknown. To investigate these questions, we employed a combined approach involving electrophysiology experiments and computational modeling. In particular, we recorded neural responses in inferotemporal (IT) and perirhinal (PRH) cortex as monkeys performed a visual target search task, and we adopted population-based read-outs to measure the amount and format of information contained in these neural populations. In Chapter 2 we report that the total amount of target match information was matched in IT and PRH, but this information was contained in a more "explicit" (i.e. linearly separable) format in PRH. These results suggest that PRH implements an "untangling" computation to reformat its inputs from IT. Consistent with this hypothesis, a simple linear-nonlinear model was sufficient to capture the transformation between the two areas. In Chapter 3, we report that the untangling computation in PRH takes time to evolve. While this type of dynamic reformatting is normally attributed to complex recurrent circuits, here we demonstrated that this phenomenon could be accounted by the same instantaneous linear-nonlinear model presented in Chapter 2. This counterintuitive finding was due to the existence of non-stationarities in the IT neural representation. Finally, in Chapter 4 we completely describe a novel set of methods that we developed and applied in Chapters 2 and 3 to quantify the task-specific signals contained in the heterogeneous neural responses in IT and PRH, and to relate these signals to measures of task performance. Together, this body of work revealed a previously unknown untangling computation in PRH during visual search, and demonstrated that a feed-forward linear-nonlinear model is sufficient to describe this computation.

Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS