Will Yandex, Cambridge, or Oxford participants be eligible for the competition?

We accept teams affiliated with Yandex, OATML Group (Oxford), and CUED Speech Group (Cambridge) to be participants in our competition. However, such participants are ineligible for prizes and will be excluded from the final leaderboard.

I don’t have a team — what should I do?

We accept individual participants as well as formed teams. If you feel you have a better chance as part of a team, you can find one in our Slack community.

Can I participate in more than one track?

Yes. Participation in any track is not mutually exclusive with participation in other tracks.

How is participation different from prize claiming?

To participate, please register, download the data and example code, produce a model and submit the results. To claim the prize for the second phase of the competition, the participants will need to submit a detailed technical report about their paper, and provide reproducible code as well as their models so they can be locally verified. The participants must also not be members of Yandex, OATML or CUED Speech Group.

Is there a paper with more details on the tasks, datasets, and the competition itself?

Of course. You can find it here on arXiv.

Are there any computational limitations?

In our competition we place a hard limitation on the computation which solutions can use. Models must run on, at most, 1xV100 GPU GPU with 32 GB of graphics memory and yield predictions within 200 ms per 1 input sample.