Prof. Kreiman gives a lecture at UC Berkeley

We now have powerful computer vision algorithms that can segment scenes, label objects, and recognize actions. It is tempting to use these algorithms as models of visual processing in biological brains. I will provide an overview of some of the successes in using neural network models to partially describe visual behavior and predict neuronal responses in the visual system of monkeys and humans, focusing on the ventral stream associated with object recognition. I will discuss passive viewing tasks as well as active visual search tasks. Next, I will provide evidence for several remaining gaps in our understanding of biological vision at both the behavioral and neuronal circuit levels, under passive and active viewing conditions. I will conclude on a more speculative note, highlighting Hilbert-like questions in the field and potential paths toward bidirectional synergies between computer vision and biological vision.