Research map / Reinforcement learning / Policy gradient and actor-critic
Accelerating Quadratic Optimization with Reinforcement Learning
Jeffrey Ichnowski, Paras Jain, Bartolomeo Stellato, Goran Banjac, Michael Luo, Francesco Borrelli, Joseph E. Gonzalez, Ion Stoica, Ken Goldberg
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 |
|---|---|---|---|
| use vector rho adaptation instead of scalar rho adaptation | RLQP · QPLIB | solve time (s): 205.372 -> 24.083 (instance 8559) | Improved |
| use general RLQP (vector) instead of OSQP heuristic | OSQP · QPLIB | solve time (s): 59.648 -> 24.083 (instance 8559) | Mixed |
| use general RLQP (vector) instead of OSQP heuristic | OSQP · Netlib LP | shifted geomean speedup: 1.30x | Improved |
| use general RLQP (vector) instead of OSQP heuristic | OSQP · Maros-Meszaros | shifted geomean runtime: 1.829 | Improved |
| train policy specific to the problem class instead of general benchmark policy | RLQP · GOMP SQP QPs | shifted geomean vs OSQP: 1.37 | Improved |
| train on low-dimensional problems only (10-50) vs full range (10-2000) (ablation) | RLQP vector policy · Random QP, dimension scaling | solve time | Hurt |
| include primal and dual residuals in policy state (Vector 1 vs 2) (ablation) | RLQP vector policy · benchmark QPs | geomean solve time | No clear effect |
| use 3 hidden layers instead of 2 (Vector 2 vs 3) (ablation) | RLQP vector policy · benchmark QPs | geomean solve time | No clear effect |
| tune training (smaller buffer, lower lr, larger batch, longer training) (ablation) | RLQP vector policy · Netlib LP | shifted geomean speedup: 1.23x -> 1.30x | Improved |
Closest papers by meaning
- Hyperparameters in Reinforcement Learning and How To Tune Them (ICML 2023 · similarity 0.88)
- Cal-QL: Calibrated Offline RL Pre-Training for Efficient Online Fine-Tuning (NeurIPS 2023 · similarity 0.88)
- Adaptive $Q$-Network: On-the-fly Target Selection for Deep Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Deep Distributed Optimization for Large-Scale Quadratic Programming (ICLR 2025 · similarity 0.87)
- Accelerated Policy Gradient: On the Convergence Rates of the Nesterov Momentum for Reinforcement Learning (ICML 2024 · similarity 0.87)
- $Q\sharp$: Provably Optimal Distributional RL for LLM Post-Training (NeurIPS 2025 · similarity 0.86)
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.
