I just wanted to have some clarification on what exactly constitutes cheating when submitting models.
- Can I use the script here to further boost my test score? https://codalab.lisn.upsaclay.fr/forums/9264/1759/
- Can we use behavioral data and experimental design? https://codalab.lisn.upsaclay.fr/forums/9264/1859/
- Can we use the single trial data? https://codalab.lisn.upsaclay.fr/forums/9264/1767/
And another question, how will this be handled if any of this constitutes cheating and a top entry used it? Will other models move up or will the leaderboard be fixed after the deadline?
Thanks for any clarification on this.
- Can I use the script here to further boost my test score? --> If you end up winning the Challenge you will need to provide a report and code relative to your models, highlighting your modeling choices. Now, how will your model (and modeling choices) contribute to the scientific community if it is a mere result of overfitting? How will the community benefit from such model? The point of the challenge is to advance modeling of the brain, and winning by overfitting defeats this purpose.
- Can we use behavioral data and experimental design? --> Yes, you can use the behavioral data. What do you mean exactly with "experimental design"?
- Can we use the single trial data? --> Yes.
The Algonauts Team
Posted by: giffordale95 @ July 24, 2023, 12:21 p.m.Is that we can use anything we can found in the NSD aws bucket to build a model?
I also found these stuff particular interesting:
- resting state, prf and floc data
- extra voxels (outside nsdgeneral) in fsaverage, func1mm preparation data, beta2 preparation
- eye tracking camera (only a few session though) and eye voxel
- demographic, session feedback excel table
Hi huze, yes, you can use the data you mentioned, as long as it doesn't consist in brain responses for the test images.
The Algonauts Team
Posted by: giffordale95 @ July 25, 2023, 10 a.m.