The authors study grouped offline contextual bandits, where several actions can be observed for each context.
Pairwise reward learning helps when context effects distort absolute scores
The authors prove that within-context reward differences cancel shared nuisance effects, but can sacrifice useful information and increase variance.
Big Tech
Junghyun Lee · Minsoo Ha · Sanghwa Kim · Yeongjong Kim · Eunjee Lee · Seiyun Shin · +1 more
Kim Jaechul Graduate School of AI, KAIST · Graduate School of AI, POSTECH · Samsung Research · Independent Researcher
Research Digest··3 min read
Lee and colleagues compare learning from individual rewards with learning from reward differences between actions shown in the same context.
Why this paper
From Samsung Research and 3 others
In one line
Pairwise loss removes nuisance-induced bias in reward learning but may increase variance, with no clear winner over pointwise loss.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§