Relative Trajectory Balance is, up to a constant, the Trust-PCL objective, and a corrected REINFORCE baseline matches RTB on the RTB paper's toy example.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Relative Trajectory Balance is equivalent to Trust-PCL
Relative Trajectory Balance is, up to a constant, the Trust-PCL objective, and a corrected REINFORCE baseline matches RTB on the RTB paper's toy example.