Keynote Speaker
Luca Benini
ETH Zürich
Title: Extreme Edge AI on Open Hardware
Abstract
Edge Artificial Intelligence (AI) is the new mega-trend, as privacy concerns and networks bandwidth/latency bottlenecks prevent cloud offloading of AI functions in many application domains, from autonomous driving to advanced prosthetics. Hence we need to push AI toward sensors and actuators. I will give an overview of recent efforts in developing systems of-on-chips based on open source hardware and capable of significant analytics and AI functions "at the extreme edge", i.e. within the limited power budget of traditional microcontrollers that can be co-located and integrated with the sensors/actuators themselves. These open, extreme edge AI platforms create an exciting playground for research and innovation.
Bio
Luca Benini holds the chair of digital Circuits and systems at ETHZ and is Full Professor at the Universita di Bologna. He received a PhD from Stanford University. In 2009-2012 he served as chief architect in STmicroelectronics France. Dr. Benini's research interests are in energy-efficient computing systems design, from embedded to high-performance. He is also active in the design ultra-low power VLSI Circuits and smart sensing micro-systems. He has published more than 1000 peer-reviewed papers and five books. He is an ERC-advanced grant winner, a Fellow of the IEEE, of the ACM and a member of the Academia Europaea. He is the recipient of the 2016 IEEE CAS Mac Van Valkenburg award and of the 2019 IEEE TCAD Donald O. Pederson Best Paper Award.
Invited Speakers
Carole-Jean Wu
Facebook AI, Arizona State University
Title: Machine Learning At Scale: Heterogeneity and Scalability Challenges for ML Systems
Abstract
Machine learning systems are being widely deployed in production datacenter infrastructure and over billions of edge devices. This talk seeks to address key system design challenges when scaling machine learning solutions to billions of people. What are key similarities and differences between cloud and edge infrastructure? The talk will conclude with open system research directions for deploying machine learning at scale.
Bio
Carole-Jean Wu is a Research Scientist at Facebook’s AI Infrastructure Research. She is also a tenured Associate Professor of CSE in Arizona State University. Carole-Jean’s research focuses in Computer and System Architectures. More recently, her research has pivoted into designing systems for machine learning. She is the leading author of “Machine Learning at Facebook: Understanding Inference at the Edge” that presents unique design challenges faced when deploying ML solutions at scale to the edge, from over billions of smartphones to Facebook’s virtual reality platforms. Carole-Jean received her Ph.D. and M.A. from Princeton and B.Sc. from Cornell.
Rune Holm
Arm
Title: Big neural networks in small spaces: Towards end-to-end optimisation for ML at the edge
Abstract
Neural networks have taken over use case after use case, from image recognition, speech recognition, image enhancement to driving cars, and show no sign of letting up. Yet so many of these use cases are done by acquiring data and sending it off to the cloud for inference. On-device ML brings unprecedented capabilities and opportunities to edge devices with improved privacy, security, and reliability. This talk explores the many aspects of system optimisation for edge ML, from training-time optimisation, to the compilation of neural networks, to the design of machine learning hardware, and looks at ways to save execution time and memory footprint while preserving accuracy.
Bio
Rune Holm has been part of the semiconductor industry for more than a decade. He started out on Mali GPUs, doing GPU microarchitecture and designing shader compilers for VLIW cores. He then moved on to research into experimental GPGPU designs and architectures targeting HPC, machine learning and computer vision. He’s currently part of the Arm Machine Learning Group, focusing on neural network accelerator architecture and compilers optimising for these designs.
Albert Cohen
Google, Paris
Title: Abstractions, Algorithms and Infrastructure for Post-Moore Optimizing Compilers
Abstract
MLIR is a recently announced open source infrastructure to accelerate innovation in machine learning (ML) and high-performance computing (HPC). It addresses the growing software and hardware fragmentation across machine learning frameworks, enabling machine learning models to be consistently represented and executed on any type of hardware. It also unifies graph representations and operators for ML and HPC. It facilitates the design and implementation of code generators, translators and optimizations at different levels of abstraction and also across application domains, hardware targets and execution environments.
We will share our vision, progress and plans in the MLIR project, zooming in on graph-level and loop nest optimization as illustrative examples.
Bio
Albert Cohen is a research scientist at Google. He worked as a research scientist at Inria from 2000 to 2018. He graduated from École Normale Supérieure de Lyon and received his PhD from the University of Versailles in 1999 (awarded two national prizes). He has also been a visiting scholar at the University of Illinois, an invited professor at Philips Research, and a visiting scientist at Facebook Artificial Intelligence Research. Albert Cohen works on parallelizing and optimizing compilers, parallel programming languages and systems, and synchronous programming for reactive control systems. He served as the general or program chair of some of the main conferences in the area and a member of the editorial board of two journals. He co-authored more than 180 peer-reviewed papers and has been the advisor for 26 PhD theses. Several research projects led by Albert Cohen resulted in effective transfer to production compilers and programming environments.