NeuralNetworkLayerTask
Design and analyze neural network layers with formal mathematical definitions, intuitive explanations, and rigorous stability analysis.
Category: Writing
Analysis: Comprehensive
Output: Multi-Tab Report
ExecutionConfig.json
UTF-8
{
"layer_name": "CustomResidual",
"forward_function_description": "y = x + f(x) where f is a MLP",
"input_shape": "[batch, d_model]",
"output_shape": "[batch, d_model]",
"parameters": ["W1", "b1", "W2", "b2"],
"activation": "relu",
"analysis_depth": "standard",
"include_lyapunov": true,
"implementation_languages": ["tensorflow.js"]
}→
SessionTask UI
Live Render
Executive Summary
Intuitive
Formal
Stability
Key Insight
The residual connection allows gradients to flow unimpeded through the identity mapping, mitigating the vanishing gradient problem in deep architectures.
Analogy: Like a bypass valve in a plumbing system that ensures water pressure remains constant even if the main filters are clogged.
Configuration Parameters
| Field | Type | Description |
|---|---|---|
layer_name * |
String | Name of the layer type (e.g., 'Attention', 'BatchNorm'). |
forward_function_description * |
String | Mathematical or natural language description of the forward pass. |
input_shape |
String | Tensor specification (e.g., '[batch, channels, h, w]'). |
output_shape |
String | Output tensor shape specification. |
parameters |
List<String> | List of learnable parameters and their intended shapes. |
activation |
String | Activation function (e.g., 'relu', 'sigmoid', 'none'). |
analysis_depth |
Enum | basic, standard, or comprehensive. |
include_lyapunov |
Boolean | Include training stability analysis (Default: true). |
include_higher_order |
Boolean | Include Hessian and curvature analysis (Default: true). |
include_lipschitz |
Boolean | Include Lipschitz continuity analysis (Default: true). |
implementation_languages |
List<String> | Target languages (e.g., 'tensorflow.js', 'python'). |