Axiomata Kan Creator — Axiomata Kan 创建器

Axiomata KAN 创建器 v1.0

Universal KAN (Kolmogorov-Arnold Network) concept creation 工具.

信息 Value Version 1.0.0 Type KAN architecture creation Architecture B-spline basis functions Requires PyTorch >= 1.9

• Purpose

Axiomata KAN 创建器 创建s KAN (Kolmogorov-Arnold Network) concepts for 代理 系统s.

KANs are neural networks that use learnable B-spline basis functions instead of fixed activation functions:

Traditional MLP: y = σ(Wx + b) — Fixed activation KAN: y = Σφᵢₙ(xᵢ) — Learnable activation

Each weight is a function (B-spline), not a scalar. This allows KANs to be more interpretable and efficient than MLPs.

• When to Use

Trigger Action "创建 a KAN" 运行 kan_创建器.py with name and 角色 "Build KAN architecture" 创建 KAN with custom layers "初始化 KAN 模型" 生成 模型 structure with B-splines "创建 KAN 流水线" Build multi-KAN 系统

• Prerequisites

Requirement Version 检查 Python >= 3.8 python3 --version PyTorch >= 1.9 python3 -c "导入 torch; print(torch.__version__)"

Note: PyTorch is required for all KAN operations (模型 creation, trAIning, inference).

• Quick 启动

4.1 创建 Basic KAN cd <技能-directory> python3 scripts/kan_创建器.py --name my_kan --角色 "监控ing"

Expected 输出:

✅ KAN 'my_kan' 创建d at scripts/my_kan/ 📋 Config: scripts/my_kan/config.json 🧠 模型: scripts/my_kan/模型s/my_kan.py

4.2 创建 KAN with Custom Parameters python3 scripts/kan_创建器.py \ --name stc_watchdog \ --角色 "emotional tension" \ --代理 morgana \ --输入-size 768 \ --输出-size 3 \ --hidden-size 32

• Architecture

5.1 KAN Layer Structure ╔═══════════════════════════════════════════════════════════╗ ║ KAN LAYER — B-Spline Trans格式化ion ║ ╠═══════════════════════════════════════════════════════════╣ ║ ║ ║ 输入: x ∈ R^输入_size ║ ║ ↓ ║ ║ B-Spline: φ(x) = ΣcᵢBᵢ(x) ║ ║ ↓ ║ ║ Learnable coefficients: cᵢ ║ ║ ↓ ║ ║ SiLU activation: σ(x) = x / (1 + e^(-x)) ║ ║ ↓ ║ ║ 输出: y ∈ R^输出_size ║ ║ ║ ╚═══════════════════════════════════════════════════════════╝

5.2 Default Architecture Parameter Default Description 输入_size 768 Embedding dimension hidden_size 32 Hidden layer width 输出_size 3 Decision dimension grid_size 5 B-spline grid points k 3 B-spline order layers [768, 32, 16, 8, 4, 3] Layer dimensions 5.3 KAN vs MLP Aspect MLP KAN Weights Scalar (fixed) Function (learnable) Activation Fixed (ReLU/sigmoid) Learnable (B-spline) Interpretability Low High TrAIning efficiency High Medium Data efficiency Medium High

• Usage

6.1 Command Reference # Basic creation python3 scripts/kan_创建器.py --name --角色 <角色>

# Full options python3 scripts/kan_创建器.py \ --name \ --角色 \ --代理 \ --输入-size \ --输出-size \ --hidden-size \ --grid-size \ --layers <列出>

6.2 Parameters Parameter Default Description --name required KAN name (used for directory/files) --角色 required KAN 角色/purpose --代理 "系统" 代理 owning the KAN --输入-size 768 输入 dimension --输出-size 3 输出 dimension --hidden-size 32 Hidden layer width --grid-size 5 B-spline grid size --k 3 B-spline order --layers auto Layer dimensions (auto-生成d if not specified) 6.3 输出 Structure / ├── config.json # KAN configuration ├── 模型s/ │ └── .py # KAN 模型 class ├── data/ │ └── trAIning/ # TrAIning data directory └── scripts/ └── trAIn.sh # TrAIning script template

• Examples

Example 1: 创建 监控ing KAN python3 scripts/kan_创建器.py \ --name stc_监控 \ --角色 "emotional tension 监控ing" \ --输出-size 3

输出:

✅ KAN 'stc_监控' 创建d 📁 scripts/stc_监控/ 📋 config.json 🧠 模型s/stc_监控.py

Example 2: 创建 Evaluation KAN python3 scripts/kan_创建器.py \ --name eval_kan \ --角色 "技能 质量 evaluation" \ --输出-size 3 \ --输入-size 768

Example 3: 创建 记录ic 验证 KAN python3 scripts/kan_创建器.py \ --name vls_kan \ --角色 "记录ic 验证" \ --代理 ezekiel \ --输出-size 3

• KAN Classes

8.1 KANLayer

Single KAN layer with B-spline basis functions:

class KANLayer(nn.模块): def __init__(self, in_features, out_features, grid_size=5, k=3): # B-spline grid: grid_size + k points # Learnable coefficients per 输出 neuron

8.2 KAN模型

Full KAN 模型 with multiple layers:

class KAN模型(nn.模块): def __init__(