AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Elastic Decision Transformer

Yueh-Hua Wu, Xiaolong Wang, Masashi Hamaya

NeurIPS 2023 · Architectures and efficiency › Attention and transformers · approach: Transformer block and depth design · 5 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
fixed history length 20 instead of variable (ablation)EDT · D4RL hopper-medium-replaynormalized return: 67.6 -> 89.0Improved
fixed history length 5 instead of variable (ablation)EDT · D4RL hopper-medium-replaynormalized return: 62.6 -> 89.0Improved
fixed history length 20 instead of variable (ablation)EDT · D4RL walker-medium-replaynormalized return: 64.5 -> 74.8Improved
fixed history length 5 instead of variable (ablation)EDT · D4RL walker-medium-replaynormalized return: 44.3 -> 74.8Improved
fixed history length 20 instead of variable (ablation)EDT · D4RL halfcheetah-mediumnormalized return: 42.0 -> 42.5No clear effect
fixed history length 20 vs variable, summed over 8 datasets (ablation)EDT · D4RL sumnormalized return: 507.6 -> 570.3Improved
fixed history length 5 vs variable, summed over 8 datasets (ablation)EDT · D4RL sumnormalized return: 483.9 -> 570.3Improved
lower expectile level alpha toward 0.5 (MSE) (ablation)EDT · Hopper medium-replaynormalized returnHurt
larger step size delta shrinking history search space (ablation)EDT · D4RL locomotionnormalized returnHurt
EDT vs DT in multi-task medium-replay locomotion (sum)EDT-1 vs DT-1 · Multi-task D4RLnormalized return: 191.3 -> 276.4Improved
EDT vs IQL in multi-task medium-replay locomotion (sum)EDT-1 vs IQL-1 · Multi-task D4RLnormalized return: 235.1 -> 276.4Improved
EDT vs DT on single-task hopper-medium-replayEDT · D4RL hopper-medium-replaynormalized return: 61.9 -> 89.0Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy