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

We tackle the Multi-task Batch Reinforcement Learning problem. Given multipledatasets collected from different tasks, we train a multi-task policy toperform well in unseen tasks sampled from the same distribution. The taskidentities of the unseen tasks are not provided. To perform well, the policymust infer the task identity from collected transitions by modelling itsdependency on states, actions and rewards. Because the different datasets mayhave state-action distributions with large divergence, the task inferencemodule can learn to ignore the rewards and spuriously correlate $\textit{only}$state-action pairs to the task identity, leading to poor test time performance.To robustify task inference, we propose a novel application of the tripletloss. To mine hard negative examples, we relabel the transitions from thetraining tasks by approximating their reward functions. When we allow furthertraining on the unseen tasks, using the trained policy as an initializationleads to significantly faster convergence compared to randomly initializedpolicies (up to $80\%$ improvement and across 5 different Mujoco taskdistributions). We name our method $\textbf{MBML}$($\textbf{M}\text{ulti-task}$ $\textbf{B}\text{atch}$ RL with$\textbf{M}\text{etric}$ $\textbf{L}\text{earning}$).

Multi-task Batch Reinforcement Learning with Metric Learning