Title: Reverse Engineering Primate Visual Object Perception
Abstract: Neuroscience is hard at work on one of our last great scientific quests — to reverse engineer the human mind and its intelligent behavior. Yet neuroscience is still in its infancy, and forward engineering approaches that aim to emulate human intelligence in artificial systems (AI) are also still in their infancy. The challenges of reverse engineering the human mind machine can only be solved by tightly coupling the efforts of brain and cognitive scientists (hypothesis generation and data acquisition), with forward engineering efforts using neurally-mechanistic computational models (hypothesis instantiation and data prediction). As this approach discovers the correct neural network models, those models will not only encapsulate our understanding of complex brain systems, they will be the basis of next-generation computing and novel brain interfaces (chemical, genetic, optical, electronic, etc.) for therapeutic and augmentation goals (e.g, brain disorders). To make this vision concrete, I will discuss one aspect of perceptual intelligence — object categorization and detection — and I will describe how work in brain science, cognitive science and computer science has converged to create deep neural networks that have recently made dramatic leaps: Not only are these neural network models reaching human performance for many images and tasks, but we have found that their internal workings largely emulate the previously mysterious internal workings of the primate ventral visual stream. Yet, our recent results show that the primate ventral visual stream still outperforms current generation artificial deep neural networks, and they point to what is importantly missing from current deep neural network models. More broadly, we believe that the community is poised to embrace a powerful new paradigm for systems neuroscience research.