Research map / Reinforcement learning / Visual and representation RL
Understanding and Improving Hyperbolic Deep Reinforcement Learning
Timo Klein, Thomas Lang, Andrii Shkabrii, Alexander Sturm, Kevin Sidak, Lukas Miklautz, Claudia Plant, Yllka Velaj, Sebastian Tschiatschek
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 changed | Where | Measured | Effect |
|---|---|---|---|
| remove RMSNorm and 1/sqrt(d) feature scaling (ablation) | Hyper++ · ProcGen PPO | test IQM: complete learning failure | Hurt |
| add learned feature scaling (ablation) | Hyper++ · ProcGen PPO | test IQM | Improved |
| use HL-Gauss categorical value loss instead of MSE (ablation) | Hyper++ · ProcGen PPO | test IQM | Improved |
| use HL-Gauss instead of C51 loss (ablation) | Hyper++ · ProcGen PPO | test IQM | Improved |
| use Hyperboloid instead of Poincare ball (ablation) | Hyper++ · ProcGen PPO | test IQM: 0.34 -> 0.40 | Improved |
| replace RMSNorm with SpectralNorm (full encoder or penultimate layer) (ablation) | Hyper++ · ProcGen PPO | test IQM: fails to learn | Hurt |
| add categorical HL-Gauss loss to Euclidean agent (ablation) | Euclidean PPO · ProcGen PPO | test IQM | Hurt |
| add RMSNorm to Euclidean agent (ablation) | Euclidean PPO · ProcGen PPO | test IQM | Improved |
| use hyperbolic Hyper++ instead of full-method Euclidean agent (ablation) | Hyper++ · ProcGen PPO | test IQM: 0.35 -> 0.40 | Improved |
| use Hyper++ instead of Hyper+S-RYM | PPG on ProcGen · ProcGen PPG | test IQM | Improved |
| use Hyper++ with Double DQN | Double DQN · Atari-5 | human-normalized score | Improved |
| Polyak vs hard target updates (ablation) | Hyper++ DDQN · NameThisGame | score | No clear effect |
Closest papers by meaning
- Stable Gradients for Stable Learning at Scale in Deep Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Hyperspherical Normalization for Scalable Deep Reinforcement Learning (ICML 2025 · similarity 0.87)
- No Representation, No Trust: Connecting Representation, Collapse, and Trust Issues in PPO (NeurIPS 2024 · similarity 0.86)
- ReMax: A Simple, Effective, and Efficient Reinforcement Learning Method for Aligning Large Language Models (ICML 2024 · similarity 0.86)
- Towards General-Purpose Model-Free Reinforcement Learning (ICLR 2025 · similarity 0.86)
- Stop Regressing: Training Value Functions via Classification for Scalable Deep RL (ICML 2024 · similarity 0.85)
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.
