Reverse Engineering Visual Intellligence

 This is a talk byJim DiCarlo from MIT at CCBM 2018 on 'Reverse Engineering Visual Intelligence'.  Jim's lab does really great work on modeling and understanding the IT cortex (responsible for object recognition).

There's a really great slide in this talk that shows off the ability of various kinds of computational models to represent specific details associated with IT cortex behavior. It's interesting that CNN AI models developed for ImageNet recognition originally surpassed the computational visual models, but then started to actually get worse even as those CNN models got better at the ImageNet Recognition task.


Popular posts from this blog

Simulating the Universe with Machine Learning

CycleGAN: a GAN architecture for learning unpaired image to image transformations

Pix2Pix: a GAN architecture for image to image transformation