Research map / Reinforcement learning / Visual and representation RL
Self-Supervised Reinforcement Learning that Transfers using Random Features
Boyuan Chen, Chuning Zhu, Pulkit Agrawal, Kaiqing Zhang, Abhishek Gupta
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 |
|---|---|---|---|
| increase random feature dimension from 128 to 2048 (ablation) | random-feature transfer method · Point | final return: 30.46 -> 34.01 | Improved |
| increase random feature dimension from 128 to 1024 (ablation) | random-feature transfer method · Metaworld Reach | final return: 765.38 -> 855.14 | Improved |
| increase random feature dimension from 128 to 512 (ablation) | random-feature transfer method · Point Perturbed | final return: 29.61 -> 33.13 | Improved |
| increase random feature dimension from 1024 to 2048 (ablation) | random-feature transfer method · Metaworld Reach | final return: 855.14 -> 830.84 | No clear effect |
| very high dimensional random features (ablation) | random-feature transfer method · Point, Point Perturbed, Metaworld Reach | adaptation speed | Hurt |
| use Gaussian matrix random features instead of NN random features (ablation) | random-feature transfer method · Point and MetaWorld Reach | reward approximation / return | Hurt |
| use second-order polynomial features instead of NN random features (ablation) | random-feature transfer method · Point and MetaWorld Reach | reward approximation / return | Mixed |
| use multi-step Q functions from random cumulants instead of dynamics model rollouts (ablation) | random-feature transfer method · novel task at test time | truncated Q approximation error | Improved |
Closest papers by meaning
- Accelerating Goal-Conditioned Reinforcement Learning Algorithms and Research (ICLR 2025 · similarity 0.91)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.90)
- Stabilizing Contrastive RL: Techniques for Robotic Goal Reaching from Offline Data (ICLR 2024 · similarity 0.89)
- Does Self-supervised Learning Really Improve Reinforcement Learning from Pixels? (NeurIPS 2022 · similarity 0.89)
- Towards General-Purpose Model-Free Reinforcement Learning (ICLR 2025 · similarity 0.88)
- 1000 Layer Networks for Self-Supervised RL: Scaling Depth Can Enable New Goal-Reaching Capabilities (NeurIPS 2025 · similarity 0.88)
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.
