Abstract: There has been tremendous progress in the development of deep convolutional neural network algorithms to address a wide variety of problems including pattern recognition tasks such as object labeling and games like chess or Go. These neural networks have been inspired by decades old research in Neuroscience elucidating the mechanisms underlying visual processing along the ventral visual stream. Despite these notable advances, human cognition still surpasses the best Artificial Intelligence algorithms to date in most problems in visual cognition. In this talk, I will outline specific examples of how advances in Neuroscience research can push the frontiers in AI. I will focus on problems like pattern completion, context reasoning, and visual attention, that require an interplay between bottom-up inputs and top-down signals that can integrate current inputs with task goals and previous knowledge. By combining behavioral measurements, neurophysiological recordings, and computational models, we can begin to decipher principles of brain computations that can be incorporated into novel biologically-inspired AI approaches.
Bio: Gabriel Kreiman is a Professor in the Department of Ophthalmology at Harvard Medical School. He is also a faculty at Children’s Hospital, the Department of Neurology at HMS, the Center for Brain Science, the Swartz Center for Theoretical Neuroscience and the Mind, Brain and Behavior Initiative at Harvard. He studied Physical Chemistry for his B.Sc. at the University of Buenos Aires, Argentina (1996). He received a M.Sc. and Ph.D. (Biology and Computational Neuroscience) from the California Institute of Technology in 2002 under Prof. Koch’s mentorship. He pursued postdoctoral work with Prof. Poggio at MIT. The Kreiman Laboratory combines computational modeling, neurophysiological recordings, and psychophysical measurements to further our understanding of the neuronal circuits and mechanisms underlying perception and cognition.