CAT-CMDP
A sequential decision algorithm proposed in the source work that, within episodic confounded MDPs, uses multi-source complementary data to identify target-domain reward and transition mechanisms, based on Gaussian dynamics and linear function approximation.
Timeline (last 90 days)
Under Gaussian dynamics and linear function approximation, the authors obtain the algorithm CAT-CMDP.
The authors establish high-probability finite-sample suboptimality guarantees, quantifying the effects of approximation error, estimation error, source-to-target shift, offline coverage and sample size.
All Facts (2)
The authors establish high-probability finite-sample suboptimality guarantees, quantifying the effects of approximation error, estimation error, source-to-target shift, offline coverage and sample size.
50%UnverifiedUnder Gaussian dynamics and linear function approximation, the authors obtain the algorithm CAT-CMDP.
50%