Derives a log-determinant CIG reward from ensemble disagreement kernel with Cholesky factorization to produce causal per-step exploration signals that scale to high-dimensional spaces.
OGBench: Benchmark- ing Offline Goal-Conditioned RL
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
CIG: Exploration via Conditional Information Gain
Derives a log-determinant CIG reward from ensemble disagreement kernel with Cholesky factorization to produce causal per-step exploration signals that scale to high-dimensional spaces.