Chips & Compilers Symposium at MLSys’21
Efficient and scalable learning systems need to take full advantage of modern hardware and runtime environments’ performance capability. The rapid advance in ML models and ML-specific hardware makes this task increasingly challenging. This symposium aims to bring together experts from the field of computer architecture and compilers, to discuss about ML-specific hardware, and how to fully use hardware specialization and compilation techniques to provide the best performance for modern machine learning workloads.
Speakers
Bill Dally
Willard R. and Inez Kerr Bell Professor at Stanford, and Chief Scientist at NVIDIA
Chris Lattner
President, Engineering and Product at SiFive
Cliff Young
Software Engineer at Google
Jacques Pienaar
Software Engineer at Google
Luis Ceze
Professor at University of Washington, Co-founder and CEO at OctoML
Mike Urbach
Founder and CEO at Alloy
Saman Amarasinghe
Professor, Department of Electrical Engineering and Computer Science at MIT
Tomas Edso
Senior Principal Design Engineer at Arm
Xiaoyong Liu
Alibaba
Zachary DeVito
Research scientist at Facebook AI Research
Organizers
Mu Li
Sr. Principal Scientist at AWS
Tianqi Chen
Assistant Professor at CMU, CTO of OctoML
Agenda
Morning Session (Pacific Time)
- 8:30–8:35
- Opening
- Organizers
- 8:35–9:30
- Logarithmic Numbers and Asynchronous Accumulators, The Future of DL Chips
- Bill Dally
- 9:30–10:00
- How Usability Improves Performance in PyTorch
- Zachary DeVito
- 10:00–10:30
Break
- 10:30–11:00
- Science to Fuel Neural Nets and TPU Design
- Cliff Young
- 11:00–11:30
- MLIR & MLIR in TensorFlow ecosystem
- Jacques Pienaar
Afternoon Session
- 13:00–13:40
- Supporting Sparse Computations in Machine Learning
- Saman Amarasinghe
- 13:40–14:10
- Applying Circuit IR Compilers and Tools (CIRCT) to ML Applications.
- Mike Urbach
- 14:10–14:30
Break
- 14:30–15:00
- Towards Automatic Scheduling for Tensorized Computation
- Tianqi Chen
- 15:00–15:30
- Endpoint AI and the Advent of the microNPU
- Tomas Edso
- 15:30–16:30
- Panel Discussion
- Chris Lattner, Luis Ceze, Cliff Young
- 16:30–17:00
- Compiler-Native Co-design for AI Heterogeneous Computing
- Xiaoyong Liu