Play research
Distillation by duplication: The importance of layer selection's itch.io pageResults
| Criteria | Rank | Score* | Raw Score |
| Generality | #9 | 2.858 | 3.500 |
| Novelty | #10 | 2.858 | 3.500 |
| Mechanistic interpretability | #12 | 2.858 | 3.500 |
| ML Safety | #14 | 1.633 | 2.000 |
| Reproducibility | #15 | 1.633 | 2.000 |
Ranked from 2 ratings. Score is adjusted from raw score by the median number of ratings per game in the jam.
Judge feedback
Judge feedback is anonymous.
- This work is quite interesting and presents a unique approach to interpreting how performance differs between student models trained from different baseline teacher layers. The visualizations are also very creative. It would be interesting to see this work expanded with more statistical analyses on additional teacher models or reinitializations to determine if the results are not just random effects (e.g. for 3,10 and 10,3). Overall, great job and I look forward to seeing this work developed further!
What are the full names of your participants?
Roksana Goworek, Paul Martin, Jonathan Frennert
What is your team name?
teamEd
What is you and your team's career stage?
UG students
Does anyone from your team want to work towards publishing this work later?
Yes
Where are you participating from?
Edinburgh
Leave a comment
Log in with itch.io to leave a comment.
Comments
No one has posted a comment yet