Aart Tells Us About Machine Learning Compilers (MLIR)

Dr. Aart Bik is a Staff Software Engineer at Google where he works on MLIR, a compiler architecture project that uses extensible compiler infrastructure alongside LLVM to build better compilers for Machine Learning.Timestamps:1:10 - An explanation of MLIR and LLVM compilers2:30 - Compilers are more than just translating high level code to assembly6:00 - Progressive Lowering in LLVM and MLIR7:40 - MLIR is largely agnostic of hardware9:00 - TPUs9:40 - AlphaZero, the ML and TPU powered chess engine, explained13:00 - The ML behind language translation16:30 - Will the future advancements in ML come hardware or from software?20:39 - Performance improvements with MLIR22:43 - Using fuzz testing to build better software including compilers31:00 - Code up a hobbyist project based on LLVM if you want to learn more about compilers32:28 - In your opinion, what is the best piece of software ever built?

Om Podcasten

We share information.