issues
search
EdoardoPona
/
predicting-inductive-biases-RL
fork of https://openreview.net/forum?id=mNtmhaDkAr - extending for inductive bias in RL
1
stars
0
forks
source link
Understanding MDL calculations
#16
Open
diogo-cruz
opened
1 year ago
diogo-cruz
commented
1 year ago
This involves:
Understanding how the MDL is being computed in both the toy and LLM tasks, in Lovering.
Understanding why our toy MDL values differ from the ones in the Lovering paper.
Understanding what changes need to be made to compute the MDL for the sentiment/summarization LLM+RL tasks.
(Optional) Understanding how the AUC metric is computed.
diogo-cruz
commented
1 year ago
As a subtask:
Implement MDL without AUC calculations
Reduce number of epochs
Reduce number of examples used
This involves: