AI Paper Atlas

Research map / Reinforcement learning / Offline RL

Uncertainty Estimation Using Riemannian Model Dynamics for Offline Reinforcement Learning

Guy Tennenholtz, Shie Mannor

NeurIPS 2022 · Reinforcement learning › Offline RL · approach: Pessimism and uncertainty in offline RL · 3 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
Use the pullback metric rather than latent ℓ2 distance (ablation)GELATO · D4RL continuous controlreturn: Hopper med: 1320 -> 1676; Walker2d med: 312 -> 1269; Hopper med-expert: 815 -> 574Mixed
Use the pullback metric rather than bootstrap ensemble uncertaintyGELATO vs MOPO · D4RL continuous controlreturn: Hopper med: 1202 -> 1676; HalfCheetah rand: 4114 -> 2560Mixed
Use the pullback metric rather than latent Euclidean distance (ablation)GELATO · Autonomous driving benchmarksreturnImproved
Use the pullback metric rather than standard uncertainty estimation methodsGELATO · Autonomous driving benchmarksreturnImproved
Use the pullback metric for uncertainty estimationGELATO · Autonomous driving benchmarksruntime: 5 times slowerHurt
Train the variance decoder in a separate second phase after calibrated Gaussian decodingGELATO variational model · D4RLconvergence speed and accuracyImproved
Use geodesic distance instead of latent Euclidean distanceGELATO distance visualization · Four Roomsdistance fidelityImproved
Use geodesic rather than Euclidean distances for the embeddingGELATO visualization · Intersectionstructure capturedImproved

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