Research map / Reinforcement learning / Visual and representation RL
RL4RLA: Teaching ML to Discover Randomized Linear Algebra Algorithms Through Curriculum Design and Graph-Based Search
Jinglong Xiong, Xiaotian Liu, Ruoxin Wang, Zihang Liu, Yefan Zhou, Yujun Yan, Yaoqing Yang
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 |
|---|---|---|---|
| add staged curriculum (full four-stage) instead of direct or partial-curriculum search (ablation) | RL4RLA on Newton Sketch · Newton Sketch | success rate: 0% -> 100% | Improved |
| skip any intermediate curriculum stage (ablation) | RL4RLA · Newton Sketch partial curricula | success rate: 100% -> 0% | Hurt |
| replace MCTS with Monte Carlo Graph Search (state merging) (ablation) | RL4RLA · Preconditioned Weighted SGD | total playouts: 34902 -> 10721 | Improved |
| replace MCTS with MCGS+UCD (ablation) | RL4RLA · Block Randomized Kaczmarz | total playouts: 66468 -> 25158 | Improved |
| replace MCTS with MCGS+UCD (ablation) | RL4RLA · Subsampled Least Square GD | total playouts: 15847 -> 5061 | Improved |
| replace MCTS with MCGS+UCD (ablation) | RL4RLA · Newton Sketch | total playouts: 2557 -> 1416 | Improved |
| use UCD instead of UCT within MCGS (ablation) | MCGS · Preconditioned Weighted SGD | total playouts: 13037 -> 10721 | Improved |
| use UCD instead of UCT within MCGS (ablation) | MCGS · Newton Sketch | total playouts: 2682 -> 1416 | Improved |
| use UCD instead of UCT within MCGS on simple targets (ablation) | MCGS · simple targets (e.g. Landweber iteration) | discovery speed | No clear effect |
| replace MCTS with MCGS on shallow targets (ablation) | RL4RLA · shallow targets (Landweber iteration) | discovery speed | No clear effect |
| replace MCTS with MCGS for wall-clock time (ablation) | RL4RLA · Subsampled Least Square GD | wall-clock time (s): 10.4 -> 8.9 | Mixed |
| MCGS success rate vs MCTS on Block Randomized Kaczmarz with UCD (ablation) | RL4RLA · Block Randomized Kaczmarz | success rate: 75% -> 75% | No clear effect |
Closest papers by meaning
- Computationally Efficient RL under Linear Bellman Completeness for Deterministic Dynamics (ICLR 2025 · similarity 0.85)
- SRL: Scaling Distributed Reinforcement Learning to Over Ten Thousand Cores (ICLR 2024 · similarity 0.84)
- Learning to Discover Iterative Spectral Algorithms (NeurIPS 2026 · similarity 0.84)
- Learning to Relax: Setting Solver Parameters Across a Sequence of Linear System Instances (ICLR 2024 · similarity 0.83)
- The Effective Horizon Explains Deep RL Performance in Stochastic Environments (ICLR 2024 · similarity 0.83)
- MATH-Beyond: A Benchmark for RL to Expand Beyond the Base Model (ICLR 2026 · similarity 0.83)
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.
