Doctoral Thesis: Language Evolution for Parallel and Scientific Computing

Friday, July 19
10:30 am - 12:30 pm

32-G449

Add to Calendar
FacebookTwitterLinkedInEmailShare

By: Valentin Churavy

Thesis Supervisor: Alan Edelman

Details

  • Date: Friday, July 19
  • Time: 10:30 am - 12:30 pm
  • Category:
  • Location: 32-G449
Additional Location Details:

Abstract:
Computer modelling is a central activity of modern science and being able to effectively take advantage
parallel hardware is crucial for many scientists. Yet parallel programming has a reputation for being hard
and many programming languages focus less on productivity and user experience and more on performance.
A programming language for scientists should also enable scientific techniques like the adjoint method.

In my talk, I will show how Julia evolved from a serial to parallel programming language, to enable large-scale scientific computer modelling. Using accelerated computing in particular, I will discuss how using array based abstraction mixed with low-level portable kernel-programming provides an effective programming environment for scientific computing. I will touch on differentiable programming as a crucial component of both the adjoint method and machine learning enabled science.

Host