[hpc-announce] [Call for AI Challenges] 2019 BenchCouncil International AI System and Algorithm Challenges (Award Presentation on Bench 19, Nov 14-16 at Denver, Colorado, USA)

[gaowanling at ict.ac.cn]

[Apologies if you receive multiple copies of this message]

                                            CALL FOR CHALLENGES

==========================================================

              2019 BenchCouncil International AI System and Algorithm Challenges

                      http://www.benchcouncil.org/competition/index.html

                           Awards: 500,000 CNY (about 70,000 US dollar)

==========================================================


Introduction
------------------------

BenchCouncil 2019 International AI system and algorithm challenges are organized by International Open Benchmark Council (BenchCouncil), and the main purpose is to advance the state-of-the-art and state-of-the-practice algorithms on different systems or architectures, i.e., RISC-V, Cambrian chip, and X86_64, and solicit new approaches to advance the state-of-the-art or state-of-the-practice algorithms. The challenge tracks use AIBench as baseline, which is publicly available from http://www.benchcouncil.org/AIBench/index.html . BenchCouncil provides the tesbed for reproducing performance numbers. The competition team can apply for nodes through http://www.benchcouncil.org/testbed/index.php .

2019 AI challenges have four tracks:
 •    International AI System Challenge based on RISC-V
 •    International AI System Challenge based on Cambrian Chip
 •    International AI System Challenge based on X86 Platform
 •    International 3D Face Recognition Algorithm Challenge


Challenges manuals
------------------------
http://www.benchcouncil.org/competition/handbook-en.pdf


Communication Tool
------------------------

The discussion groups are hosted on Xinxiu---a dedicated communication tool for science and education.
 Discussion group site: https://app.ic3i.com/org/benchcouncil/en/


 Important Dates
------------------------
Now the AI challenge is beginning!
Registration deadline: Sep. 15, 2019, anywhere on earth
Performance numbers finalized: October 1, 2019, anywhere on earth
The code should be submitted to BenchHub for reproducing performance numbers and code review. 
Submission site: http://125.39.136.212:8090
Preliminary paper version submitted: October 15, 2019, anywhere on earth
Paper submission：
https://easychair.org/conferences/?conf=competition2019
Camera-ready version submitted: November 10, 2019


Awards
------------------------
Special Award (Only one): 100,000 CNY
The First Prize: 
•    30,000 CNY (one for every track)
The Second Prize: 
•    20,000 CNY (two for every track)
The Third Prize: 
•    10,000 CNY (three for every track)


Awards Presentation
------------------------
The award presentation is on Bench 19 conference ( http://www.benchcouncil.org/bench19/index.html ), which will be held on Nov 14-16 at Denver, Colorado, USA.

Every team should submit their paper to BenchCouncil International Symposium on Benchmarking, Measuring and Optimizing (Bench 19). The award-winners must submit a paper to Bench 19 conference and give a presentation.
Bench19 Submission Site: https://easychair.org/conferences/?conf=competition2019

Award committees
•    Lizy John (University of Texas at Austin)
•    D. K. Panda (OSU)
•    Geoffrey Fox (Indianan University)
•    Wanling Gao (ICT, Chinese Academy of Sciences)
•    XIaoyi Lu (OSU)
•    Jianfeng Zhan (ICT, Chinese Academy of Sciences)


AI Challenge Tracks
------------------------
(1) International AI System Competition based on RISC-V
Goal
•    The implementation and optimization of CNN-based image classification task on RISC-V, using Cifar-10 dataset and ResNet-50 model
Targets
•    Implement the forward calculation stage
•    Minimize external dependences (e.g., OpenMP, Boost)
•    Guarantee the original model accuracy (deviation<0.05%)
Metrics
•    Maximize the execution performance (number of instructions)
•    Minimize the binary file, e.g., compiled executable file

(2) International AI System Competition based on Cambrian Chip
Goal
•    The implementation and optimization of CNN-based image classification task on Cambrian, using Cifar-10 dataset and ResNet-50 model
Target
•    Implement the forward calculation stage
•    Guarantee the original model accuracy (deviation<0.05%)
Metric
•    Maximize the execution performance---the shorter the prediction time on test data, the better the performance

(3) International AI System Competition based on X86 Platform
Goal
•    The implementation and optimization of matrix decomposition based collaborative filtering task on X86 platform, using MovieLens dataset and ALS-WR algorithm
Target
•    Implement ALS-WR training algorithm
•    Can use external libraries supported by the platform
Metric
•    Maximize the execution performance-reduce training time (30 rounds)

(4) International 3D Face Recognition Algorithm Competition
Goal
•    Innovative algorithm for 3D Face Recognition
Targets
•    The competitors need to submit the model file and test file
•    Description file, source code
•    External data for training is allowed, but need description
Metrics
•    ROC and AUC value