LAB

CMC Microsystems Atlas 800 AI Competition

Important Deadlines:

  • July 18, 2022:  Start date.
  • September 29, 2022: Entry Deadline.
  • January 15, 2023: Final Submission Deadline.
  • February 15, 2023: Winners announced

Description

CMC Microsystems is pleased to organize an AI competition focusing on the development, acceleration and deployment of deep learning models targeting the Atlas 800 training cluster. The goal of this competition is to enable Canadian academic researchers/developers to utilize Huawei’s new compute resources and expand the range of model compatibility to a dedicated hardware architecture.

What is Atlas 800?

Featuring Huawei Ascend AI processor (NPU) and CANN (computing architecture for neural networks), Atlas 800 is a high-performance compute server fully dedicated to accelerate AI training and inference, with support for popular ML frameworks (PyTorch, TensorFlow, MindSpore). 

Eligible students can apply here:

Contact Us

Dr. Yassine Hariri
Senior Staff Scientist
AI/ML and Embedded Systems

Eligibility

The Competition is open to legal residents of Canada who: (i) have reached the age of majority in their jurisdiction of residence; and (ii) are Canadian university students with a valid CMC Microsystems subscription. A free trial subscription is also available for eligible Professors / Research Supervisors.

Further, you must submit an entry in one of the following Competition Tracks (each a “Track”). Your entry must meet the requirements of each track to be considered valid. Entrants can submit an entry in multiple Tracks.

Terms and Conditions

By participating in the Competition, each entrant agrees to abide by and be bound by the Official Rules.

Track Submission Details and Prizes

Track 1: SOTA Model Migration

A list of candidate models with GPU code open-sourced. Participants may also propose their model of choice; The participants will make necessary adaptations of the training script to train the model on the NPU server with target performance, and open source their implementation. See a list of Recommended Models for SOTA Model Migration.

There are five awards available to be won, ranging from $2000 to $5000 CAD, which will be allocated as follows:

  • 1st place: $5000
  • 2nd place: $4000
  • 3rd place: $3500
  • 4th place: $2500
  • 5th place: $2000

A panel of judges will evaluate each entry based on the following criteria:

  • CR1: Performance of the model running on the NPU including the accuracy and training speed (Weight=70%)
  • CR2: Quality of the code and documentation (Weight=15%)
  • CR3: Ease of use (Weight =15%)

Participant's Github URL, including training log, Readme, codes that can run on the NPU server, open-sourced: e.g., under MIT or Apache license.  

Track 2: Research on NPU

Participants train their model on the NPU and publish paper(s) with their codes open-sourced.

There are three awards available to be won for up to $5000 CAD per paper, which will be allocated as follows:

  • 1st place: $5000
  • 2nd place: $4000
  • 3rd place: $3500

Paper submission and code; For papers not accepted, they will be evaluated based on paper and result quality. A panel of judges will evaluate each entry based on the following criteria:

  • CR1: Demo and presentation (Weight =50%)
  • CR2: Performance of the model running on the NPU including the accuracy and training speed (Weight =20%)
  • CR3: Quality of the code and documentation (Weight =15%)
  • CR4: Ease of use (Weight =15%)
  • For accepted papers: Copy of the paper; Code Github URL; code can run on NPU server, open-sourced: e.g. under MIT or Apache license.
  • For submitted papers: Live presentation and demo

Track 3: Innovative Application on NPU

Participants build innovative application(s) on the NPU.

  • E.g., applications on Robotics/Autonomous driving. See application ideas.
  • Cloud AI: Atlas 800 Server can provide web service [ref]
  • Edge AI (robotics, autonomous driving, etc): In this case, you can use Atlas 200 DK, the edge AI device, instead Atlas 800.
  • Example: Robot car; Drone

There are three awards available to be won, ranging from $3500 to $5000 CAD, which will be allocated as follows:

  • 1st place: $5000
  • 2nd place: $4000
  • 3rd place: $3500

A panel of judges will evaluate each entry based on the following criteria:

  • CR1: innovation, performance, technical difficulty, usability (Weight =70%)
  • CR2: Quality of the code and documentation (Weight =15%)
  • CR3: Ease of use (Weight =15%)

Code Github URL, report, demo video

Track 4: Team Category 

Participants in this track are required to form a team with at least one member from an equity-seeking group. The teams can apply to one or multiple tracks.

There are two awards available to be won, ranging from $3000 to $5000 CAD per application, which will be allocated as follows:

  • 1st place: $5000
  • 2nd place: $3000

Depends on the selected track.

Depends on the selected track. See submission details above for each track.

Scroll to Top
X
Skip to content