菜单
Analyze Spiking Neural Network (SNN) papers, extract technical patterns from knowledge graph, and identify reusable research methodologies for neuromorphic computing.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/hiyenwong/ai_collection --skill spiking-neural-network-analysis命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Graph-native Python reimplementation of the Information Dynamics of Music (IDyOM) model that represents predictive memories as explicit graph objects for musical expectation modeling and network analysis.
Physics-aware end-to-end deep reinforcement learning methodology for quadcopter control with actuator dynamics modeling.
Reinforced Dreamer methodology for asymmetric reinforcement learning using latent guidance to improve world model representations and behaviors in model-based RL.
基于 SOC 职业分类
正在显示 SKILL.md
| name | spiking-neural-network-analysis |
| version | v1.0.0 |
| last_updated | 2026-04-06T00:00:00.000Z |
| description | Analyze Spiking Neural Network (SNN) papers, extract technical patterns from knowledge graph, and identify reusable research methodologies for neuromorphic computing. |
Analyze SNN research papers from knowledge graph and extract reusable technical patterns.
exec - Run kg_tool and sqlite3 commandsread - Read skill docs and paper contentwrite - Save analysis reportsUser Request → Identify Task Type
├── Search Papers → Search kg.db for SNN papers
├── Analyze Patterns → Extract technical patterns
├── Knowledge Graph → Run PageRank/Louvain/Similarity
└── Generate Report → Create analysis summary
# Search knowledge graph
sqlite3 /Users/hiyenwong/wiki/kg.db \
"SELECT id, name FROM kg_entities WHERE name LIKE '%spiking%' OR name LIKE '%neuromorphic%'"
# Get paper properties (keywords, topics, category)
sqlite3 /Users/hiyenwong/wiki/kg.db \
"SELECT id, name, properties FROM kg_entities WHERE entity_type='paper' AND name LIKE '%spiking%'"
# PageRank
/Users/hiyenwong/.openclaw/workspace/scripts/kg_tool/target/release/kg_tool pagerank --limit 10
# Community detection
/Users/hiyenwong/.openclaw/workspace/scripts/kg_tool/target/release/kg_tool communities --limit 10
# Search
/Users/hiyenwong/.openclaw/workspace/scripts/kg_tool/target/release/kg_tool search --query "spiking neural" --limit 10
Common SNN patterns:
User: "分析 SNN 论文"
Execute:
1. Search "spiking neural" papers
2. Get Top 5 paper properties
3. Run pagerank
4. Extract patterns
5. Generate report
User: "从 SNN 论文中提取技术模式"
Execute:
1. List all SNN papers
2. Parse keywords from properties
3. Identify common patterns
4. Create pattern summary
User: "找出与 SNN 相关的量子计算论文"
Execute:
1. Search "spiking" AND "quantum"
2. Find intersection entities
3. Analyze hybrid approaches
| Paper | Category | Pattern |
|---|---|---|
| Energy-Efficient SNN | Medical AI | energy-efficient |
| Brain-Inspired Computing | Neuromorphic | brain-inspired |
| PSPM (Pre-Synaptic Pool) | Supervised Learning | forward-forward |
arXiv: 2605.07207 — Luu et al. (2026), IEEE Signal Processing Letters
A critical SNN deployment pattern: converting direct-coded SNNs (floating-point inputs) to event-based representations (TTFS) for neuromorphic hardware.
Core approach: Self-Knowledge Distillation (SKD) — use the pretrained direct-coded SNN as its own teacher to guide event-based finetuning via KL-divergence regularization.
Key theorem: Cross-domain accuracy gap bounded by KL divergence between teacher/student output distributions + TV distance between input distributions.
Performance: SKD recovers 45-51pp on CIFAR-10 (vs naive TSF's 30-49pp), consistently outperforms across 9 architectures.
Theoretical limits: ~60% of input entropy is lost in TTFS encoding — even optimal transfer cannot match direct-coded teacher. Expect 15-20pp ceiling gap.
Pitfalls:
qutrit-entropy-neural-estimation skill.Core innovation: Couples pretrained EfficientNet encoder with CoLaNET spiking classifier via rate-coding. Trains SNN using local biologically-inspired learning rules, bypassing end-to-end backpropagation.
Performance: 99.09% on 64-class ImageNet benchmark — on par with conventional deep networks.
Why it matters: Demonstrates that pretrained ANN features + local plasticity can replace full backpropagation, enabling biologically-plausible yet high-performance SNNs.
Pattern: Pretrained Encoder → Rate Coding → Local Learning SNN — modular, energy-efficient, no gradient transport.
Core innovation: Automated black-box assessment of spiking neuron models using Izhikevich firing pattern classification. Encodes 20 canonical patterns as objective functions for optimization.
Implementation: Python/PyTorch/Norse compatible. Treats neuron models as black boxes — no analytical modeling required.
Why it matters: Provides empirical, systematic way to quantify "how biological" a neuron model is. Useful for comparing neuromorphic hardware designs.
Pattern: Model → Pattern Encoding → Black-Box Optimization → Plausibility Score
SNNs can be augmented with quantum computing for energy-efficient high-performance tasks:
quantum-neuromorphic-computing skill for the full methodologybrain-connectivity-analysis - Brain network analysisquantum-neural-hybrid - Quantum-classical NNquantized-snn-hardware-optimization - Hardware acceleration/Users/hiyenwong/wiki/kg.db/Users/hiyenwong/.openclaw/workspace/scripts/kg_tool/target/release/kg_toolmemory/2026-04-06.mdreferences/functional-ensembles-1fc-groups.md - 1FC (First-order Functional Connectivity) groups framework from arXiv 2606.00073. Documents computational units in deep SNNs through inter-layer functional connectivity analysis, including methodology, experimental findings, and applications for network debugging, architecture optimization, and transfer learning.