Research map / Reinforcement learning / Visual and representation RL
Symbolic Distillation for Learned TCP Congestion Control
S P Sharan, Wenqing Zheng, Kuo-Feng Hsu, Jiarong Xing, Ang Chen, Zhangyang Wang
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 context-specific branches and a branch decider to SymbolicPCC (ablation) | SymbolicPCC · Mininet lossy network | Lossy mean throughput: 28.40 -> 28.55 | Improved |
| Add context-specific branches and a branch decider to SymbolicPCC (ablation) | SymbolicPCC · Mininet lossy network | Lossy total throughput: 738.46 -> 742.46 | Improved |
| Add context-specific branches and a branch decider to SymbolicPCC (ablation) | SymbolicPCC · Mininet lossy network | Lossy mean squared error to optimal: 7.29 -> 4.14 | Improved |
| Add context-specific branches and a branch decider to SymbolicPCC (ablation) | SymbolicPCC · Mininet oscillating network | Oscillating mean squared error to optimal: 85.03 -> 43.83 | Improved |
| Add context-specific branches and a branch decider to SymbolicPCC (ablation) | SymbolicPCC · CC agent inference | Runtime (μs): 23 -> 37 | Hurt |
| Distill Aurora into a symbolic policy | Aurora · Mininet lossy network | Lossy mean squared error to optimal: 26.29 -> 7.29 | Improved |
| Distill Aurora into a symbolic policy | Aurora · CC agent inference | Runtime (μs): 864 -> 23 | Improved |
| Prune Aurora by 50% | Aurora · Mininet lossy network | Lossy mean throughput: 27.55 -> 27.03 | Hurt |
| Prune Aurora by 80% | Aurora · Mininet lossy network | Lossy mean throughput: 27.55 -> 26.42 | Hurt |
| Prune Aurora by 95% | Aurora · Mininet lossy network | Lossy mean throughput: 27.55 -> 25.97 | Hurt |
| Quantize Aurora | Aurora · Mininet lossy network | Lossy mean throughput: 27.55 -> 22.54 | Hurt |
| Distill Aurora into a 50%-size black-box model, then distill that model symbolically | Aurora · PCC-RL | Average return: 832 -> 641 -> 687 | Mixed |
Closest papers by meaning
- Efficient Reinforcement Learning by Discovering Neural Pathways (NeurIPS 2024 · similarity 0.83)
- End-to-End Neuro-Symbolic Reinforcement Learning with Textual Explanations (ICML 2024 · similarity 0.81)
- Verifying Neural Networks with Reinforcement Learning (NeurIPS 2026 · similarity 0.81)
- BlendRL: A Framework for Merging Symbolic and Neural Policy Learning (ICLR 2025 · similarity 0.81)
- The Ladder in Chaos: Improving Policy Learning by Harnessing the Parameter Evolving Path in A Low-dimensional Space (NeurIPS 2024 · similarity 0.81)
- Offline Behavior Distillation (NeurIPS 2024 · similarity 0.81)
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.
