AutoML Decathlon 2022
Diverse Tasks, Modern Methods, and Efficiency at Scale
Competition Rules
The competition rules, meant to ensure fair and accurate evaluation of submissions, are specified below:
- Submissions must be free and reproducible, and the winning solutions must be made public after the competition ends. You may use paid tools to develop your submissions, however after your solution is made public, reproducing its results on the benchmarking tasks must be free for users. We will not require you to open-source your methods. For example, you could release an API that allows users to run and experiment with the tool only on the tasks from the competition.
- Participants must not extract any information about the ground truth labels for the test task datasets through any means or attempt any prohibited behavior listed in the following section that may be considered as cheating.
- Submissions may reuse open-source code only with proper attribution.
- Submissions will only be considered valid if they satisfy the imposed computational constriants for evaluation, detailed on the CodaLab page.
- Participants are free (and in fact encouraged) to use additional resources or data from related tasks beyond those provided by the organizers.
We strictly prohibit any form of cheating, including:
- Copying existing code or repositories without proper attribution
- Two teams having similar or identical solutions
- Attempting to access test datasets
- Any other actions we deem to be inappropriate