The model and the brain

People sometimes ask whether deep learning is a good model of the brain. I think that’s the wrong question.

The brain isn’t trying to be efficient in the way a neural network is. It’s doing something stranger — running on noisy hardware, under metabolic constraints, across decades of experience. The gap between the two is where the interesting questions live.

I’ve stopped being bothered by the fact that my models don’t map cleanly onto biology. The mismatch is data.