UC Berkeley

The human cerebral cortex comprises many functionally distinct areas that represent different information about the world. It has been challenging to map these areas efficiently. In this dissertation, I present a new approach that addresses this problem. In chapter one, I present a novel voxelwise encoding model based on Tikhonov regression. I discuss the theoretical basis for Tikhonov regression, demonstrate a computationally efficient method for its application, and show several examples of how Tikhonov regression can improve predictive models for fMRI data. I also show that many earlier studies have implicitly used Tikhonov regression by linearly transforming the regressors before performing ridge regression. In chapter two, I present a critique of an alternative method used to study brain representations called representational similarity analysis. I show that this method makes strong assumptions about the relationship between representational models and brain responses. I also show that representational similarity analysis can lead to incorrect conclusions when used to compare representational models. In chapter three, I present a rich paradigm for efficient non-invasive functional brain mapping. In this paradigm, subjects watch interesting short films while their brain activity is measured. Multiple feature spaces are used to model the brain responses to the short films. Each feature space constitutes a hypothesis about the type of representations that might be important for brain regions involved in watching, listening, and understanding the short films. The novel voxelwise encoding model developed in chapter one is then used to find the most predictive feature spaces across the cortical surface and also to recover maps that capture how the individual feature spaces are represented within cortical regions. The results suggest a high degree of homogeneous selectivity for feature spaces across large regions of the cortical surface within individual subjects. These patterns are highly consistent across all subjects. Finally, I explore the functional organization of the middle temporal cortex and show that the visual feature spaces can capture novel functional subdivisions in this region.