%0 Conference Paper %1 SziSze11 %A Szita, I. %A Szepesvári, Cs. %B COLT %D 2011 %K adaptivity, bounds, complexity finite-sample learning, model off-policy offline regularization, reinforcement selection, theory %P 739--772 %T Agnostic KWIK learning and efficient approximate reinforcement learning %X A popular approach in reinforcement learning is to use a model-based algorithm, i.e., an algorithm that utilizes a model learner to learn an approximate model to the environment. It has been shown such a model-based learner is efficient if the model learner is efficient in the so-called ``knows what it knows'' (KWIK) framework. A major limitation of the standard KWIK framework is that, by its very definition, it covers only the case when the (model) learner can represent the actual environment with no errors. In this paper, we introduce the agnostic KWIK learning model, where we relax this assumption by allowing nonzero approximation errors. We show that with the new definition that an efficient model learner still leads to an efficient reinforcement learning algorithm. At the same time, though, we find that learning within the new framework can be substantially slower as compared to the standard framework, even in the case of simple learning problems.

@inproceedings{SziSze11, abstract = {A popular approach in reinforcement learning is to use a model-based algorithm, i.e., an algorithm that utilizes a model learner to learn an approximate model to the environment. It has been shown such a model-based learner is efficient if the model learner is efficient in the so-called ``knows what it knows'' (KWIK) framework. A major limitation of the standard KWIK framework is that, by its very definition, it covers only the case when the (model) learner can represent the actual environment with no errors. In this paper, we introduce the agnostic KWIK learning model, where we relax this assumption by allowing nonzero approximation errors. We show that with the new definition that an efficient model learner still leads to an efficient reinforcement learning algorithm. At the same time, though, we find that learning within the new framework can be substantially slower as compared to the standard framework, even in the case of simple learning problems.}, added-at = {2020-03-17T03:03:01.000+0100}, author = {Szita, I. and Szepesv{\'a}ri, {Cs}.}, biburl = {https://www.bibsonomy.org/bibtex/233cf8be2daa54c1ecfc28ee26bde0c79/csaba}, booktitle = {COLT}, date-added = {2011-07-03 20:59:06 -0600}, date-modified = {2012-06-03 14:10:27 -0600}, interhash = {5b6bdd47b4be7f8fc2cb3156097ee9b4}, intrahash = {33cf8be2daa54c1ecfc28ee26bde0c79}, keywords = {adaptivity, bounds, complexity finite-sample learning, model off-policy offline regularization, reinforcement selection, theory}, month = {July}, pages = {739--772}, pdf = {papers/agnosticKwik.pdf}, timestamp = {2020-03-17T03:03:01.000+0100}, title = {Agnostic {KWIK} learning and efficient approximate reinforcement learning}, year = 2011 }