A Computational Explanation for Domain Specificity in the Human Visual System

Explore a computational explanation for domain specificity in the human visual system in this 35-minute talk by Katharina Dobs from MIT. Delve into recent research testing whether the segregation of face and object perception in primate brains emerges naturally from task optimization. Learn about experiments with artificial neural networks trained on face and object recognition, revealing how network size affects task performance and pathway segregation. Discover insights into why domain-specific organization may be an effective design strategy for brains, reflecting computational optimization over development and evolution. Gain understanding of the speaker's research bridging machine learning advances with human behavioral and neural data to provide a precise account of visual recognition processes. Cover topics including functional specialization in the human visual system, convolutional neural networks as models for human vision, and experiments investigating task-specificity in neural networks.

The problem vision needs to solve
Functional specialization in the human visual system
The rise of convolutional neural networks (CNN) in computer science
CNNs as model for the human visual system
Do face and object tasks require distinct computations?
Can representations be learned to support both tasks?
Research questions
Lesion experiments in the dual-task network
Does a larger system segregate both tasks?
At which processing stage does task-specificity arise?
Do any two tasks require distinct computations?
Do dual-task networks better mimic human behavion?