Presentation
AutoDL: Automated Tools for Fast Development of Deep Learning Networks and Accelerators
DescriptionRecent breakthroughs in machine learning (ML) algorithms, especially deep neural networks (DNNs), have fueled an increasing research interest in efficient DNN solutions, aiming to bring powerful yet power-hungry DNNs into resource-constrained edge devices. Hardware-aware DNN design has emerged as one of the most promising techniques thanks to its outstanding potential. However, developing these DNN models requires cross-disciplinary expertise in the algorithm, hardware architecture, and device-specific compilation, forming a significant entry barrier for many developers. On the other hand, from the hardware side, designing domain-specific architectures for DNNs, i.e., DNN accelerators, is also prohibitively challenging. It can take a large team of hardware experts for months to design a single DNN accelerator because of the numerous design choices spanning dataflows, processing elements, memory hierarchy, and more. Moreover, it is increasingly evident that optimal DNN accelerators require full-stack co-design and co-search across the entire compute stack, from algorithmic to hardware optimizations. Unfortunately, the full potential of such codesign and co-search advancements has not been fully harnessed, largely limiting the advance and innovation of efficient DNN solutions and DNN accelerators.
The tutorial aims to educate the research community on the challenges and recent advances in automated tools for developing efficient DNN solutions and DNN accelerators. Our tutorial features high-profile speakers from both academia and industry.
Agenda:
0. Opening (5 min):
Speaker: Prof. Yingyan (Celine) Lin - Georgia Institute of Technology
1. Keynote 1 (30 min):
Speaker: Prof. Vijay Janapa Reddi - Harvard University
Title: Architecture 2.0: Why Computer Architects Need a Data-Centric AI Gymnasium
2. Invited talk (20 min):
Speaker: Yonggan Fu - Georgia Institute of Technology
Title: Algorithm Accelerator Co-search towards Fast Development and More Efficient AI Accelerators
3. Keynote 2 (30 min):
Speaker: Dr. Amir Yazdanbakhsh - Google
Title: Breaking Barriers: Embracing Machine Learning for the Next Generation of Domain Specific Accelerators
The tutorial aims to educate the research community on the challenges and recent advances in automated tools for developing efficient DNN solutions and DNN accelerators. Our tutorial features high-profile speakers from both academia and industry.
Agenda:
0. Opening (5 min):
Speaker: Prof. Yingyan (Celine) Lin - Georgia Institute of Technology
1. Keynote 1 (30 min):
Speaker: Prof. Vijay Janapa Reddi - Harvard University
Title: Architecture 2.0: Why Computer Architects Need a Data-Centric AI Gymnasium
2. Invited talk (20 min):
Speaker: Yonggan Fu - Georgia Institute of Technology
Title: Algorithm Accelerator Co-search towards Fast Development and More Efficient AI Accelerators
3. Keynote 2 (30 min):
Speaker: Dr. Amir Yazdanbakhsh - Google
Title: Breaking Barriers: Embracing Machine Learning for the Next Generation of Domain Specific Accelerators
Event Type
Tutorial
TimeMonday, July 10th10:30am - 12:00pm PDT
Location3004, 3rd Floor
AI
