Skip to content
Back to skills

Synaptic Weight Distributions Plasticity Geometry

ASecurity

Synaptic Weight Distributions and Plasticity Geometry

  • 3 stars
  • 0 votes
  • 0 copies
  • 8 views
  • Added September 11, 2026
toolspythongo

Works with

  • cli

Security analysis

A100/100

Scanned September 11, 2026

npx -y skills add hiyenwong/ai_collection --skill synaptic-weight-distributions-plasticity-geometry --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Synaptic Weight Distributions Plasticity Geometry?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Synaptic Weight Distributions Plasticity Geometry
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/hiyenwong-synaptic-weight-distributions-plasticity-geometry/badge)](https://www.skillsdirectory.com/skills/hiyenwong-synaptic-weight-distributions-plasticity-geometry)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: synaptic-weight-distributions-plasticity-geometry
description: Synaptic Weight Distributions and Plasticity Geometry
---

# Synaptic Weight Distributions and Plasticity Geometry

## Description

使用镜像下降理论分析突触可塑性的几何性质,发现实验观察到的对数正态突触权重分布与非欧几里得几何一致,挑战了标准梯度下降假设。

## Activation Keywords

- synaptic weight distribution
- plasticity geometry
- mirror descent
- log-normal distribution
- non-euclidean gradient
- synaptic plasticity theory

## Tools Used

- `read` - 读取突触权重数据
- `exec` - 运行 Python 分析脚本
- `web_fetch` - 获取论文详细内容

## Instructions for Agents

### 1. 理解核心问题

**传统假设的局限:**
- 计算神经科学常用梯度下降研究突触可塑性
- 梯度下降假设欧几里得距离(欧氏几何)
- 但生物学不一定使用欧几里得几何

**关键发现:**
- 突触权重分布取决于可塑性的几何性质
- 实验观察到的对数正态分布与非欧几何一致
- 标准梯度下降(欧氏几何)无法解释这些分布

### 2. 理论框架

**镜像下降(Mirror Descent):**
```python
import numpy as np

class MirrorDescent:
    """
    镜像下降框架 - 统一不同几何的梯度下降
    
    关键组件:
    - 势函数 φ: 定义几何
    - 梯度更新在镜像空间进行
    """
    def __init__(self, potential_function='euclidean'):
        self.potential = potential_function
    
    def mirror_map(self, w):
        """
        镜像映射:权重 → 梯度空间
        不同势函数定义不同几何
        """
        if self.potential == 'euclidean':
            return w  # 欧氏几何
        elif self.potential == 'log':
            return np.log(w)  # 对数几何
        elif self.potential == 'entropy':
            return np.log(w / (1 - w))  # 熵几何
    
    def inverse_mirror_map(self, theta):
        """
        逆镜像映射:梯度空间 → 权重
        """
        if self.potential == 'euclidean':
            return theta
        elif self.potential == 'log':
            return np.exp(theta)
        elif self.potential == 'entropy':
            return 1 / (1 + np.exp(-theta))
    
    def update(self, w, gradient, lr=0.01):
        """
        镜像下降更新规则
        """
        theta = self.mirror_map(w)
        theta_new = theta - lr * gradient
        return self.inverse_mirror_map(theta_new)
```

### 3. 权重分布预测

```python
def predict_weight_distribution(geometry='euclidean', n_synapses=10000):
    """
    根据几何预测突触权重分布
    
    参数:
    - geometry: 'euclidean', 'log', 'entropy'
    - n_synapses: 模拟的突触数量
    
    返回:
    - weights: 权重分布
    """
    if geometry == 'euclidean':
        # 欧氏几何 → 正态分布
        weights = np.abs(np.random.randn(n_synapses))
    elif geometry == 'log':
        # 对数几何 → 对数正态分布
        weights = np.random.lognormal(mean=0, sigma=1, size=n_synapses)
    elif geometry == 'entropy':
        # 熵几何 → Beta 分布
        weights = np.random.beta(0.5, 0.5, size=n_synapses)
    
    return weights

def analyze_distribution(weights):
    """
    分析权重分布,判断几何类型
    """
    from scipy import stats
    
    # 检验对数正态性
    log_weights = np.log(weights[weights > 0])
    _, p_lognormal = stats.normaltest(log_weights)
    
    # 检验正态性
    _, p_normal = stats.normaltest(weights)
    
    if p_lognormal > 0.05:
        return "Consistent with log geometry (log-normal distribution)"
    elif p_normal > 0.05:
        return "Consistent with Euclidean geometry (normal distribution)"
    else:
        return "Consistent with other non-Euclidean geometry"
```

### 4. 实验验证方法

**比较学习前后的权重分布:**
1. 记录初始突触权重分布
2. 进行学习任务
3. 记录学习后权重分布
4. 分析分布变化推断几何类型

## Examples

### Example 1: 分析突触权重分布

```
User: 如何判断突触可塑性使用什么几何?

Agent: 使用 synaptic-weight-distributions-plasticity-geometry 技能:

1. 收集突触权重数据
2. 分析分布形状:
   - 正态分布 → 欧氏几何
   - 对数正态分布 → 对数几何
   - 其他分布 → 其他非欧几何

arXiv:2305.19394 发现:
- 多个脑区观察到的对数正态分布
- 与非欧几何一致,不支持欧氏几何假设
- 暗示生物学使用非欧几里得距离度量突触变化
```

### Example 2: 镜像下降实现

```
User: 如何实现非欧几何的学习规则?

Agent: 使用镜像下降框架:

# 欧氏几何(标准梯度下降)
w_new = w - lr * gradient

# 对数几何
w_new = w * exp(-lr * gradient / w)

# 熵几何
w_new = w / (w + (1-w) * exp(-lr * gradient))

不同几何导致不同的权重分布和动力学。
选择正确的几何对模拟生物学习至关重要。
```

## Source

- **arXiv:** 2305.19394
- **效用:** 0.93
- **标题:** Synaptic Weight Distributions Depend on the Geometry of Plasticity
- **会议:** ICLR 2024

## Key Findings

1. **几何决定分布** - 突触权重分布由可塑性几何决定
2. **实验不一致** - 对数正态分布与欧氏几何不匹配
3. **非欧几何** - 生物学可能使用非欧几里得距离
4. **可验证性** - 通过学习前后分布变化可实验测试

## Related Skills

- `heterogeneous-synaptic-dynamics` - 异质突触动力学
- `neuromodulated-synaptic-plasticity` - 神经调制突触可塑性
- `sparse-gradient-plasticity` - 稀疏梯度可塑性

## References

- Pogodin et al. (2024) - 原始论文
- Beck & Teboulle (2003) - 镜像下降理论
- Buzsáki & Mizuseki (2014) - 对数正态脑动力学

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in tools →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.