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

There is an increasing interest in learning reward functions that model humanpreferences. However, many frameworks use blackbox learning methods that, whileexpressive, are difficult to interpret. We propose and evaluate a novelapproach for learning expressive and interpretable reward functions frompreferences using Differentiable Decision Trees (DDTs). Our experiments acrossseveral domains, including CartPole, Visual Gridworld environments and Atarigames, provide evidence that that the tree structure of our learned rewardfunction is useful in determining the extent to which the reward function isaligned with human preferences. We provide experimental evidence that rewardDDTs can achieve competitive performance when compared with larger capacitydeep neural network reward functions. We also observe that the choice betweensoft and hard (argmax) output of reward DDT reveals a tension between wantinghighly shaped rewards to ensure good RL performance, while also wantingsimpler, more interpretable rewards.

Can Differentiable Decision Trees Learn Interpretable Reward Functions?