Do I need to submit model weights?

Hi @dd-mike,

I’d like to know whether I need to submit model weights for phase 2 submission.

Thank you.

Hi @separate. You do not need to submit your model weights, but do make sure that someone independently running the code in your “End Phase 1 Submission” (matching, descriptor) will be able to replicate your submission. As a reminder, that submission should include, among other things:

The series of commands, in order, that would get a reasonably experienced and savvy user from your code to a finished submission.

Hope this helps!

Thanks for the reply, @dd-mike.

However, I think not submitting fixed model weights before 10/19 will cause some confusion or unfair situations (like training further for 1 week after 10/19).

Of course replicating solution process will be applied, but it’ll be more secure if you could make sure that model is fixed after 10/19.

Especially, submitting model weights before 10/19 will prevent the situation like Phase 2 Leaderboard result is changed because of some teams keep training after 10/19.

If score difference between two teams in Phase 2 are very small, training one week more will directly impact the results. And organizers will not be able to identify who illegally trained more, because replicating submissions can only verify that they can get comparable results using the given training code.

Therefore I think both 1.replicating solutions, 2.Make sure that model is fixed after 10/19
are important to make competition fairer.

I think there are 3 possible solutions,
1.submitting model weights before 10/19
2.phase 2 starts directly from 10/19
3.phase 1 ends at 10/26 and phase 2 starts from 10/26

Please consider, thank you.

Hi @separate,

Thank you for sharing your concerns. We do not plan to change any rules or deadlines at this time, based on our assessment of competition integrity and disruptiveness of any changes. Rest assured that the organizers are very much aware of these issues from this and previous competitions, and the competition has been designed with these concerns in mind.

Thanks for your answer, @dd-mike