Research map / Reinforcement learning / Policy gradient and actor-critic
Error Amplification Limits ANN-to-SNN Conversion in Continuous Control
Zijie Xu, Zihan Huang, Yiting Dong, Kang Chen, Wenxuan Liu, Zhaofei Yu
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 CRPI to IF conversion (T=16) | converted DDPG · MuJoCo | average performance ratio: 82.57% -> 95.41% | Improved |
| add CRPI to IF conversion (T=8) | converted TD3 · MuJoCo | average performance ratio: 64.71% -> 72.26% | Improved |
| add CRPI to IF conversion (T=32) | converted SAC · MuJoCo | average performance ratio: 97.98% -> 99.70% | Improved |
| add CRPI to MT neurons (T=4) | converted TD3 · MuJoCo | average performance ratio: 97.93% -> 98.09% | No clear effect |
| add CRPI to IF neurons (T=32) | DrQ-v2 SNN · DMC visual | average performance ratio: 40.77% -> 74.19% | Improved |
| add CRPI to IF neurons (T=64) | DrQ-v2 SNN · DMC visual | average performance ratio: 69.60% -> 91.84% | Improved |
| add CRPI to SNM (T=8) | DrQ-v2 SNN · DMC visual | average performance ratio: 88.72% -> 96.27% | Improved |
| add CRPI to MT (T=2) | DrQ-v2 SNN · DMC visual | average performance ratio: 94.85% -> 101.30% | Improved |
| increase alpha in CRPI (ablation) | CRPI · MuJoCo TD3 IF T=16 | cross-step error correlation | Improved |
| replace ANN policy states with SNN-induced trajectory (ablation) | TD3 ANN policy (IF T=8) · Ant-v4 reward decomposition | return: 6505.26 -> 4051.77 | Hurt |
| replace ANN policy with SNN policy on ANN trajectory (ablation) | TD3 (IF T=8) · Ant-v4 reward decomposition | return: 6505.26 -> 6216.71 | No clear effect |
| use CRPI-converted SNN instead of directly trained Spiking-WM | SNN on visual control · DMC visual | average performance ratio: 68.54% -> 74.19% | Improved |
Closest papers by meaning
- Adaptive Calibration: A Unified Conversion Framework of Spiking Neural Networks (AAAI 2025 · similarity 0.88)
- Efficient ANN-SNN Conversion with Error Compensation Learning (ICML 2025 · similarity 0.88)
- Proxy Target: Bridging the Gap Between Discrete Spiking Neural Networks and Continuous Control (NeurIPS 2025 · similarity 0.88)
- Temporal Misalignment in ANN-SNN Conversion and its Mitigation via Probabilistic Spiking Neurons (ICML 2025 · similarity 0.86)
- Faster and Stronger: When ANN-SNN Conversion Meets Parallel Spiking Calculation (ICML 2025 · similarity 0.86)
- Adaptive Surrogate Gradients for Sequential Reinforcement Learning in Spiking Neural Networks (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.
