Apurva Ratan Murty
Postdoc, Massachusetts Institute of Technology
The last quarter century has provided extensive evidence that some regions of the human cortex are selectively engaged in processing a single specific domain of information - from faces, places, and bodies to music, language, and other people’s thoughts. This work dovetails with earlier theories in cognitive science highlighting domain specificity in human cognition, development, and evolution. But despite decades of work on these regions, we still do not have a computationally precise understanding of what visual information encoded in these regions, and many questions still remain unanswered about even the clearest cases of specificity in the brain.
In this talk I will describe a series of experiments that attempt to move us towards a computationally precise account of visual processing in the human brain. I will first critically evaluate the evidence for cortical selectivity in the human ventral visual cortex. For instance, many theories around category selectivity remain vulnerable to empirical refutation. Given these critical shortcomings, is the theory even true? To address these issues, I will show some recent efforts to reverse-engineer these regions by constructing deep artificial neural network (ANN)-based encoding models that predict the observed response to novel images, outperforming descriptive models and even professors! I will then further demonstrate how we can apply these models to make ever stronger inferences about theories of category selectivity which points the way for future research characterizing the functional organization of the human brain with unprecedented computational precision. In the remaining part of my talk, I will demonstrate how computational models can help bring computational precision to classic theories of vision and be used to revel a finer-grained organization of the human brain than known previously. Together, the experiments set the stage for a research program that will bring us to closer to a computationally precise account of primate visual representations.
Computational models of category-selective brain regions enable high-throughput tests of selectivity
Integrative Benchmarking to Advance Neurally Mechanistic Models of Human Intelligence
Visual experience is not necessary for the development of face-selectivity in the lateral fusiform gyrus