菜单
Functional Ensembles as Units of Computation in Deep Spiking Networks. 1FC (first-order functionally-connected) ensembles framework for analyzing information encoding in SNNs through rare coordinated firing events.
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 functional-ensembles-deep-spiking-networks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 | functional-ensembles-deep-spiking-networks |
| description | Functional Ensembles as Units of Computation in Deep Spiking Networks. 1FC (first-order functionally-connected) ensembles framework for analyzing information encoding in SNNs through rare coordinated firing events. |
| version | 1.0.0 |
| author | Aditi Aravind, Konstantinos Ladakis, Mario Alexios Savaglio, Stelios M. Smirnakis, Maria Papadopouli |
| arxiv_id | 2606.00073 |
| category | spiking-neural-networks |
| activation_keywords | ["functional ensemble","1FC ensemble","spiking neural network","rare event coding","cofiring analysis","adversarial robustness","functional connectivity","deep SNN"] |
| created | 2026-06-05T00:00:00.000Z |
首次提出 1FC (first-order functionally-connected) ensemble 概念，用于分析深度脉冲神经网络中的信息编码机制。核心发现：信息编码集中在稀有的高协同发放事件中，而非持续性活动。
突破性发现：
First-order Functionally-Connected Group:
For neuron i in layer L:
1FC(i) = {j in layer L-1 : corr(spike_i, spike_j) > threshold}
Aggregate cofiring predicts downstream response:
Response = ReLU-like(Σ spikes_in_1FC - threshold)
Gain ∝ ensemble_size
关键发现：
High information content ⊂ Rare high-coordination events
# 计算 pairwise correlation
def compute_1FC_groups(spike_trains_L, spike_trains_L_minus_1, threshold=0.05):
"""
spike_trains: binary arrays (n_neurons, T)
"""
correlations = np.corrcoef(spike_trains_L, spike_trains_L_minus_1)
# Statistical significance test
significant_connections = correlations > threshold
# Form 1FC groups
1FC_groups = {}
for i in (n_neurons_L):
1FC_groups[i] = np.where(significant_connections[i])[]
1FC_groups
def analyze_cofiring_events(1FC_groups, spike_trains):
"""检测高协同发放事件"""
# Aggregate cofiring for each neuron
cofiring_strength = []
for i, ensemble in enumerate(1FC_groups.items()):
ensemble_spikes = spike_trains[ensemble].sum(axis=0)
neuron_spike = spike_trains[i]
# Cofiring during neuron's spike
cofiring_when_spike = ensemble_spikes[neuron_spike > 0]
cofiring_strength.append(cofiring_when_spike.mean())
# Identify rare high-coordination events
threshold = np.percentile(cofiring_strength, 95)
high_cofiring_events = cofiring_strength > threshold
return high_cofiring_events
def measure_information_encoding(spike_trains, labels, high_cofiring_events):
"""量化稀有事件中的信息编码"""
# Compare class encoding during high vs low cofiring
spikes_high_cofiring = spike_trains[high_cofiring_events]
spikes_low_cofiring = spike_trains[~high_cofiring_events]
# Mutual information with class labels
MI_high = mutual_info_class(spikes_high_cofiring, labels)
MI_low = mutual_info_class(spikes_low_cofiring, labels)
print(f"Information during high cofiring: {MI_high}")
print(f"Information during low cofiring: {MI_low}")
print(f"Ratio: {MI_high / MI_low}") # >> 1 typically
适用场景：
训练 SNN
# Spiking ResNet architecture
model = SpikingResNet(num_layers=5)
model.train_on_dataset(images, labels)
提取 spike trains
# Record all layer spike trains during inference
spike_recordings = {}
for layer_idx in range(num_layers):
spike_recordings[layer_idx] = model.get_layer_spikes(layer_idx)
形成 1FC groups
# Compute 1FC for each layer
1FC_groups = {}
for L in range(1, num_layers):
1FC_groups[L] = compute_1FC_groups(
spike_recordings[L],
spike_recordings[L-1]
)
分析 cofiring patterns
# Detect high-coordination events
high_cofiring = analyze_cofiring_events(1FC_groups, spike_recordings)
# Measure information encoding
info_encoding = measure_information_encoding(
spike_recordings,
test_labels,
high_cofiring
)
对抗扰动诊断
# Test adversarial robustness
adversarial_inputs = generate_adversarial(model, test_images)
adversarial_spikes = model.get_all_spikes(adversarial_inputs)
# Compare 1FC structure
1FC_adversarial = compute_1FC_groups(adversarial_spikes)
# Detect disrupted layers
disrupted_layers = compare_1FC_structure(1FC_groups, 1FC_adversarial)
深度 SNN 的功能连接模式与生物皮层观察到的原则一致
1FC ensemble 集体发放 → ReLU-like 下游响应，gain 与 size 相关
稀有高协同事件集中了大部分信息编码能力
学习过程塑造功能连接结构，weight permutation 破坏这种结构
允许在特定节点和通路进行精细粒度诊断
| Approach | Scale | Temporal | Interpretability | Biological Plausibility |
|---|---|---|---|---|
| 1FC Ensemble | Fine | High | Excellent | High |
| Simple Correlation | Fine | Low | Medium | Medium |
| PCA Clustering | Coarse | None | Low | Low |
| Graph Neural | Coarse | None | Medium | Medium |
扩展到 2FC, 3FC (higher-order functional connectivity)
分析 1FC ensemble 的时序演化
比较不同 SNN architecture 的 1FC patterns
与实际皮层神经记录数据对比
基于诊断结果设计 targeted fine-tuning
Created: 2026-06-05 Source: arXiv:2606.00073