Julia (workshop)
summary
Claims Julia generates PTX GPU code so kernels compile from high-level Julia, uses dual-number automatic differentiation with zero runtime overhead, and offers DSLs/macros plus packages JuMP, Dagger, Knet and Flux for optimization, parallelism and deep learning. Demonstrations include style-transfer and a diabetic retinopathy classifier trained on low-cost fundus images. At scale, Celeste rebuilds the Sloan Digital Sky Survey catalog (tens of terabytes) in minutes on a supercomputer, reaching ~1.5 petaflops and reporting ~1000× speedups over typical single-threaded workflows.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 0:45Why Julia
- 3:20Machine Code
- 4:00Why Machine Code
- 4:40Vector Addition
- 6:00Vector Addition Kernel
- 7:00Generic Code
- 9:00Auto Diff
- 11:00Toolkit
- 13:00DomainSpecific Languages
- 14:00Flux
- 16:00Chaining
- 18:00Taking a picture
- 19:00Prisma
- 20:00Deep Neural Network
- 21:00Perceptual Loss Functions
- 23:00Other projects
- 24:00Bayesian inference
related talks
43mJulia: A Fresh Approach to Machine Learning
01h 27mIntroduction to Julia for Scientific Computing and Data Science
01h 28mA Tour of Julia
01h 06mJulia: Fast Performance, Distributed Computing & Multiple Dispatch
01h 07mJulia Tutorial
01h 37mTech Tutorial: Use Julia for your next scientific computing job!
42mWhy I Use Julia For Quantum Physics
02h 00mIntroduction to Julia