English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

In the predict-then-optimize framework, the objective is to train apredictive model, mapping from environment features to parameters of anoptimization problem, which maximizes decision quality when the optimization issubsequently solved. Recent work on decision-focused learning shows thatembedding the optimization problem in the training pipeline can improvedecision quality and help generalize better to unseen tasks compared to relyingon an intermediate loss function for evaluating prediction quality. We studythe predict-then-optimize framework in the context of sequential decisionproblems (formulated as MDPs) that are solved via reinforcement learning. Inparticular, we are given environment features and a set of trajectories fromtraining MDPs, which we use to train a predictive model that generalizes tounseen test MDPs without trajectories. Two significant computational challengesarise in applying decision-focused learning to MDPs: (i) large state and actionspaces make it infeasible for existing techniques to differentiate through MDPproblems, and (ii) the high-dimensional policy space, as parameterized by aneural network, makes differentiating through a policy expensive. We resolvethe first challenge by sampling provably unbiased derivatives to approximateand differentiate through optimality conditions, and the second challenge byusing a low-rank approximation to the high-dimensional sample-basedderivatives. We implement both Bellman--based and policy gradient--baseddecision-focused learning on three different MDP problems with missingparameters, and show that decision-focused learning performs better ingeneralization to unseen tasks.

Learning MDPs from Features: Predict-Then-Optimize for Sequential Decision Problems by Reinforcement Learning