Definition
A neural network stacks layers of simple mathematical units (neurons) that transform inputs through weighted sums and nonlinear activations.
During training, weights adjust to minimize error on examples — the network learns representations useful for classification, generation, or prediction.
In simple terms
Imagine a series of filters in a photo app: each layer detects edges, then shapes, then objects. A neural network learns those filters automatically from data instead of hand-coding them.
Where you see it
- Image classifiers, speech recognizers, and LLMs are all neural networks.
- Recommendation systems use neural nets for ranking.
- Somali NLP models use neural encoders for text classification.
How it works
1.Forward pass
Input flows through layers; each neuron computes a weighted sum plus activation.
2.Loss calculation
Compare output to the correct answer; measure how wrong the prediction is.
3.Backpropagation
Compute gradients and update weights to reduce loss.
4.Repeat
Many epochs over the dataset until performance plateaus.
Why it matters
- Neural networks are the foundation of modern AI — from tiny classifiers to trillion-parameter LLMs.
Often confused
Neural networks mimic the brain accurately.
They are inspired by biology but mathematically are matrix operations optimized with calculus — useful analogy, not literal simulation.