Stephen talked about automatic differentiation (autodiff) techniques including finite differencing, forward-mode autodiff, and reverse-mode autodiff, the last of which is also commonly known as “backpropagation”.

The talk ended with discussing how checkpoints can save memory in reverse-mode autodiff, optimally trading off between space usage and computational time.

Previous: Auction design

Brian gave an introduction to auction design, a key topic in algorithmic game theory.

continue reading ❯