Information
[2023/02/03] Notification of recalculation of final evaluation score
From 9:30 on Monday, February 6th, we will replace the benchmark function for the final evaluation and recalculate the final evaluation score of the files you have submitted so far.No action is required from you for this support, but please note that you will not be able to post new files while the score of your own submitted file is being recalculated.In addition, there is no change in the provisional evaluation score.
[2023/02/02] Apology and notification to all participants
We will extend the qualifying period of this contest and replace the benchmark function for the final evaluation.In addition, the qualifying period after the extension will be until 2/9 24:00.The contest should have ended at 24:00 on February 2nd, but due to a mistake in setting the deadline, it ended at 0:00 on February 2nd. It is a response to what happened.At around 9:35 on February 2nd, we returned the contest to its original state and resumed the contest. In order to maintain fairness, we decided to take the above measures.As soon as the replacement of the benchmark function for the final evaluation is completed, all past submission files except for error submissions will be recalculated sequentially and the final evaluation value will be updated.
We apologize for any inconvenience this may cause.There is no problem in submitting new files during the extended qualifying period.
[2022/12/23] We have started accepting submissions.
[2022/12/23] Deleted the item "Assessment of the size and execution speed" in the evaluation.
[2022/12/19] Updated schedule.
Objective
While early social implementation of artificial intelligence technology is expected, technology that accelerates the introduction of artificial intelligence technology is crucial. When introducing artificial intelligence technology there is an issue in that it takes a tremendous amount of time to tune establishment and studying of artificial intelligence models. In order to solve this problem, NEDO has a project called “Development of integration technology that becomes the core of the next-generation artificial intelligence and robots/ Research and development that expands the areas of application of artificial intelligence technology”: As a common fundamental technology that promotes early social implementation of artificial intelligence technology, the project develops technology that shortens the time to tune establishment and studying of the artificial intelligence models as well as an artificial intelligence acceleration modules using the technology.
Specifically, technologies such as exploration for Hyperparameter Optimization (HPO hereafter) and Neutral Architecture Search (NAS hereafter) become important as a tuning task in practical settings. To solve these issues we have developed an HPO Library “aiaccel” which optimizes hyperparameters in a short time for the research and development theme “Establishment of cyber physical value chain and AI introduction acceleration technology.” This Library is published as open-source software which can accommodate a newly developed HPO module. The Library can also utilize the AI Bridging Cloud Infrastructure (ABCI hereafter) which is owned by the National Institute of Advanced Industrial Science and Technology (AIST) as the development environment.
In order to further promote this project it is essential to induce technological development in a competitive environment which is joined by a wide range of participants including universities, corporations and overseas researchers. Therefore, we decided to hold an AI introduction acceleration module competition which will induce the creation of AI introduction acceleration modules that display excellent individual performance and versatility.
For the inaugural year we are holding a competition that assesses the performance of HPO modules to optimize hyperparameters, such as learning rate. Please take this wonderful opportunity to try using the ABCI, a world’s largest computing infrastructure for artificial intelligence processing.
Overview
The competition consists of the first and final rounds. Ten qualifying teams which passed the first round can proceed to the final match.
First-Round | Final Round | |
Period | December 16, 2022 to February 2, 2023. | Scheduled to start at the middle of February 2023. |
Task | Optimization of HPO module | Optimization of HPO module (Deep Learning) |
Requirement | Developing a module that works with aiaccel. | Developing a module that works with aiaccel. |
Eligibility | Must be able to participate in the final round. | Must add and publish the open-source license to the source code if awarded with a prize. Before the start of the competition personal identification and documents required to use the ABCI must be submitted(※). |
Test Environment | Implemented with a computational resource owned by the individual participants. | The optimization is tested using the ABCI’s computational resource through SIGNATE by posting the HPO module. |
Assessment Environment | Assessed with SIGNATE by posting the HPO module. | Assessed with SIGNATE by posting the HPO module. |
Assessment Method | Optimization accuracy is assessed using multiple benchmark functions. | Optimization accuracy of hyperparameters of deep learning is assessed. |
Leaderboard Update Frequency | Real time | Once a week. |
Prize | Top ten teams with high accuracy: Eligibility to enter the final round. | Quantitative evaluation: ¥700,000 × 5 teams Qualitative evaluation: ¥250,000 × 2 teams |
Tasks for the First Round
The participants will develop an HPO module that complies with the IF specifications* regulated by the HPO Library “aiaccel”. The performance will be assessed using multiple benchmarks functions.
*See Data tab for details.
Optimization image
※Note:Terms of agreement in document required to use the ABCI in the final round.
- You must meet the following requirements: You must not be in violation of the Foreign Exchange and Foreign Trade Act (No. 228 of 1949), government and ministerial ordinances and notifications related to security export control based on the Act, security export control regulation held by the National Institute of Advanced Industrial Science and Technology, and other regulations held by the National Institute of Advanced Industrial Science and Technology based on the said regulation, and you must fall under any of the following:
- A Japanese citizen who resides in Japan.
- A Japanese citizen who works for a diplomatic mission abroad.
- An alien who works for an office in Japan.
- An alien with over six months having passed since entering Japan.
- Being able to provide the following personal information on all team members:
- First and last name.
- The name of the institution or corporation the participant belongs to or its nationality.
- The name of the department of the above institution/corporation.
- Email address.
- Length of stay in Japan.
- Organization abroad the participant belongs to.
- A copy of personal identification with photo (personal identification with photo means a document that we can determine that the description written on the document exists and the corporation the user belongs to guarantees the user’s identification to the third party), or a personal identification without photo issued by the corporation the user belongs to and a copy of the driver’s license with photo.
- The participant must not fall under any of the following:
- In an employment contract with foreign corporation and government.
- Obtaining or promised to obtain a profit of more than 25% of the annual income from the foreign government.
- Receiving specific instructions or requests from a foreign government regarding behavior while in Japan. (Further details will be published on the website as soon as determined.)
Final Round
Ten teams who passed the first round advance to the final round. In the final round the optimization of hyperparameters in deep learning will be assessed. We are considering awarding a certain amount of ABCI points to the individuals who have participated in the final round.