graph TD
    A[Neural Networks] --> B[Architecture]
    A --> C[Training]
    A --> D[Activation Functions]
    B --> B1[Input Layer]
    B --> B2[Hidden Layers]
    B --> B3[Output Layer]
    C --> C1[Forward Pass]
    C --> C2[Loss Calculation]
    C --> C3[Backpropagation]
    C --> C4[Weight Update]
    D --> D1[Sigmoid]
    D --> D2[ReLU]

flowchart LR
    A[Input Data] --> B[Forward Pass]
    B --> C[Calculate Loss]
    C --> D{Loss acceptable?}
    D -->|No| E[Backpropagation]
    E --> F[Update Weights]
    F --> B
    D -->|Yes| G[Model Ready]

graph LR
    subgraph Input Layer
        I1[x1] & I2[x2]
    end
    subgraph Hidden Layer
        H1[h1] & H2[h2] & H3[h3]
    end
    subgraph Output
        O1[y]
    end
    I1 & I2 --> H1 & H2 & H3
    H1 & H2 & H3 --> O1

graph TD
    A[Activation Functions] --> B[Sigmoid]
    A --> C[ReLU]
    B --> B1["Range: 0 to 1"]
    B --> B2["Use: Output layer"]
    B --> B3["Problem: Vanishing gradient"]
    C --> C1["Range: 0 to infinity"]
    C --> C2["Use: Hidden layers"]
    C --> C3["Problem: Dead neurons"]

sequenceDiagram
    participant D as Data
    participant N as Network
    participant L as Loss Function
    participant O as Optimizer
    D->>N: Forward pass
    N->>L: Predictions
    L->>L: Calculate error
    L->>N: Gradients (backprop)
    N->>O: Current weights + gradients
    O->>N: Updated weights

stateDiagram-v2
    [*] --> Untrained
    Untrained --> Training: Start training
    Training --> Evaluating: Each epoch
    Evaluating --> Training: Loss too high
    Evaluating --> Trained: Loss acceptable
    Trained --> Deployed: Deploy
    Deployed --> Training: Retrain

# Visual Concept Maps: [Topic]

## Overview Map
[Concept hierarchy - always include this one]

## [Diagram Type 2 title]
[Most relevant additional diagram]

## [Diagram Type 3 title]
[Second most relevant]

## Key Relationships Summary
- [Concept A] depends on [Concept B] because...
- [Concept C] is an alternative to [Concept D] when...
- [Process X] feeds into [Process Y] via...

## Concept Coverage
- Concepts in diagrams: [N] / [N] from inventory
- Concepts not diagrammed: [list] (with reason: "too granular" or "no visual relationship")

graph LR
    A[Linear Algebra] --> B[Neural Network Basics]
    A --> C[Gradient Descent]
    B --> D[Backpropagation]
    C --> D
    D --> E[Training Loop]
    E --> F[PyTorch Implementation]

quadrantChart
    title Concept Difficulty vs Importance
    x-axis Low Difficulty --> High Difficulty
    y-axis Low Importance --> High Importance
    Neuron anatomy: [0.3, 0.7]
    Backpropagation: [0.8, 0.9]
    Activation functions: [0.5, 0.8]
    Learning rate tuning: [0.6, 0.7]

graph LR
    subgraph Before
        B1["Neural network = black box"]
        B2["Training = magic"]
    end
    subgraph After
        A1["Neural network = layers of math functions"]
        A2["Training = iterative error minimization"]
    end
    B1 -.->|"this lecture"| A1
    B2 -.->|"this lecture"| A2