Neural-Network Basics
Neural-Network Basics
Fact-check status
Key jargon
| Term | Plain-language meaning |
|---|---|
| Neuron | A small computation that combines inputs, weights, a bias, and an activation. S1 |
| Weight | A learned number controlling how strongly one signal affects another. S1 |
| Loss function | A score describing how wrong a prediction is for the training objective. S1 |
| Backpropagation | The procedure that computes how each parameter contributed to the loss. S1 |
Key concepts
- A neural network is a differentiable function with learned parameters, not a digital brain. S1
- Training repeatedly predicts, measures error, computes gradients, and updates weights. S1
Concept map
flowchart LR
A["Input features"] --> B["Weighted layers"]
B --> C["Prediction and loss"]
C --> D["Backpropagation updates weights"]Diagram semantics verified against S1 and S2 on 2026-09-02.
Mental model
A neural network is a parameterized function. Inputs are converted to numbers, passed through layers, and transformed into outputs. Weights store learned numerical relationships; activations introduce nonlinear behavior; a loss function measures error; backpropagation computes how parameters contributed to that error; an optimizer updates them. S1
Training is repeated approximation—not a database write of one fact into one address. S1
Vocabulary
| Term | Practical meaning |
|---|---|
| Parameter | Learned number affecting the function S1 |
| Hyperparameter | Human-selected setting such as learning rate or batch size S1 |
| Epoch | One pass over a training dataset S1 |
| Overfitting | Memorizing training-specific patterns that fail to generalize S1 |
| Regularization | Technique discouraging brittle or overly complex fits S1 |
| Gradient | Direction and sensitivity of loss change S1 |
Misconceptions
- More parameters do not guarantee better task performance. S1
- Low training loss does not prove useful generalization. S1
- A model’s internal representation is not a human-readable knowledge graph. S1
Exercise
Draw a tiny network with three inputs, one hidden layer, and two outputs. Label which values come from data, which are learned, and which are chosen by the builder.