On-device Intelligence Workshop

March 4th, 2020 in conjunction with MLSys (March 2–4)

Austin Convention Center | Austin, Texas

AI has the potential to transform almost everything around us. It can change the way humans interact with the world by making the objects around them “smart” — capable of constantly learning, adapting, and providing proactive assistance. The beginnings of this trend can already be seen in the new capabilities coming to smartphones (speech assistant, camera night mode) as well as the new class of “smart” devices such as smart watches, smart thermostats, and so on. However, these “smart” devices run much of the computation on the cloud (or a remote host) — costing them transmission power and response latency as well as causing potential privacy concerns. This limits their ability to provide a compelling user experience and realize the true potential of an “AI everywhere” world.

This workshop seeks to accelerate the transition towards a truly “smart” world where the AI capabilities permeate to all devices and sensors. The workshop will focus on how to distribute the AI capabilities across the whole system stack and co-design of edge device capabilities and AI algorithms. It will bring together researchers and practitioners with diverse backgrounds to cover the whole stack from application domains such as computer vision and speech, to the AI and machine learning algorithms that enable them, to the SoC/chip architecture that run them, and finally to the circuits, sensors, and memory technologies needed to build these devices.

Organizing Committee

  • Vikas Chandra, Facebook (Program co-chair)
  • Pete Warden, Google (Program co-chair)
  • Yingyan Lin, Rice University (General co-chair)
  • Ganesh Venkatesh, Facebook (General co-chair)
  • Ariya Rastrow, Amazon
  • Raziel Alvarez, Google
  • Song Han, MIT
  • Greg Diamos, Landing.AI
  • Hernan Badino, Facebook Reality Labs

Program Committee

  • Yingyan Lin, Rice University
  • Ariya Rastrow, Amazon
  • Raziel Alvarez, Google
  • Claudionor Coelho, Google
  • Jan Kautz, Nvidia Research
  • Zhangyang Wang, TAMU
  • Yiran Chen, Duke University
  • Richard Baraniuk, Rice University
  • Hernan Badino, Facebook Reality Labs
  • Kiran Somasundaram, Facebook Reality Labs
  • Ziyun Li, Facebook Reality Labs
  • Christian Fuegen, Facebook AI
  • Meng Li, Facebook
  • Liangzhen Lai, Facebook
  • Yu-Hsin Chen, Facebook


WednesdayMar. 4

9:00 - 9:15 am

  • Opening Remarks | Vikas Chandra

9:15 - 10:15 am

  • Keynote I | Blaise Aguera Y Arcas

10:30 - 11:00 am

  • Imitation Learning from Observation | Prof. Peter Stone

11:00 - 12:00 pm

  • Accepted Papers | (see below)

2:00 - 2:45 pm

  • Keynote II | Prof. Diana Marculescu

2:45 - 3:15 pm

  • How to evaluate Deep Learning Accelerators? | Prof. Vivienne Sze

3:15 - 3:45 pm

  • AutoML for On-device Vision | Mingxing Tan

3:45 - 4:15 pm

  • TF-lite on Microcontrollers | Nat Jeffries

4:15 - 4:30 pm

  • Closing Remarks | Pete Warden

4:30 - 5:00 pm

  • Poster Session | (see below)


Accepted Papers

Efficient Neural Network Specialization on FPGA with Once-for-all Network
Han Cai, Chuang Gan, Tianzhe Wang, Zhekai Zhang, Song Han

Accelerator-aware Neural Network Design using AutoML
Suyog Gupta, Berkin Akin

Compressing Language Models using Doped Kronecker Products
Urmish Thakker, Paul Whatamough, Matthew Mattina, Jesse Beu

Optimizing Speech Recognition for the Edge
Yuan Shangguan, Jian Li, Qiao Liang, Raziel Alvarez, Ian McGraw

Federated Optimization in Heterogeneous Networks
Tian Li, Anit Kumar Sahu, Manzil Zaheer, Maziar Sanjabi, Ameet Talwalkar, Virginia Smith

Lightweight Convolutional Representations for On-Device Natural Language Processing
Shrey Desai, Geoffrey Goh, Arun Babu, Ahmed Aly

Call for Papers

We invite participation in the On-device Intelligence Workshop being held in conjunction with Machine Learning and Systems 2020 on March 4th, 2020 in Austin, Texas. Topics of interest are anything related to enabling smart devices including but not limited to the following:

  • Novel on-device capabilities for vision, speech, and natural language processing.
  • Distributing AI capabilities across the whole system stack from data capture at the edge to the cloud instead of performing all the compute in the cloud.
  • Machine learning for system tasks such as compression, scheduling, and caching.
  • On-device privacy-preserving learning.
  • Efficient machine learning models for edge devices.
  • Dynamic neural networks such as early termination and mixture-of-experts.
  • Platform-aware model optimization.
  • Efficient hardware accelerator design.
  • Tools for architecture modeling, design space exploration, and algorithm mapping.
  • Efficient model execution on edge devices such as scheduling and tiling.
  • Emerging technologies such as near-sensor, -memory, or -storage computing.

Authors are encouraged to submit original research (including those already available as preprint), initial findings, and insights from research-in-progress or position papers on the above topics. The program committee will select submissions based on a combination of novelty, insightful or thought-provoking observations, and relevance to the workshop.

Submission Instructions

  • Submissions are recommended to be up to four pages long plus an extra page for references or citations.
  • Submissions must follow the format outlined for MLSys 2020
  • Submissions do not have to be anonymized
  • Apply here

Accepted Submissions

Accepted papers are expected to have at least one author to present the paper at the workshop. The presentation will be recorded and made available online to make the workshop accessible to those unable to attend. Each accepted submission will be a full presentation or a NeurIPS-style five-minute “spotlight” presentation plus a poster.

Important Dates

  • Paper submission deadline: January 22nd, 2020 at 5:00pm PST.
  • Decision notification: January 27th, 2020
  • Workshop: March 4, 2020