Zubnet AILearnWiki › Feature
Fundamentals

Feature

Learned Representation, Activation
A pattern or concept that a neural network learns to detect in its input. In vision, early-layer features are edges and textures; later-layer features are object parts and whole objects. In language models, features range from simple (the letter "a," a specific syntax pattern) to abstract (the concept of sarcasm, a particular reasoning strategy). Features are represented as activation patterns across neurons.

Why it matters

Features are what models actually learn — not individual facts but patterns that generalize. A model doesn't memorize "cats have fur"; it learns a feature detector for fur-like textures that activates for cats, dogs, and teddy bears. Understanding features helps explain model behavior: why it generalizes (features transfer), why it fails (wrong feature activated), and how to improve it (expose it to more diverse features).

Deep Dive

The term "feature" has different meanings depending on context. In classical ML, features are hand-engineered input variables (height, weight, age). In deep learning, features are learned representations in hidden layers — the model discovers useful patterns on its own. This shift from hand-engineered to learned features is the core innovation of deep learning and why it outperforms classical ML on complex tasks like vision and language.

Hierarchical Features

Deep networks learn hierarchical features: each layer builds on the previous one. In a vision model: layer 1 detects edges, layer 2 combines edges into textures and corners, layer 3 combines textures into object parts (eyes, wheels), layer 4 combines parts into objects (faces, cars). This hierarchy emerges automatically from training — no one programs it. The same hierarchical feature learning happens in language models, from character patterns to syntax to semantics to reasoning.

Feature Visualization

Researchers visualize features to understand what models learn. For vision models, you can generate images that maximally activate a specific neuron or direction, revealing what pattern it detects. For language models, you can find the text examples that most activate a specific feature direction. Anthropic's research has visualized features in Claude, finding interpretable concepts like "Golden Gate Bridge," "code bugs," "deception," and "French language" encoded as specific directions in the model's activation space.

Related Concepts

← All Terms
← Existential Risk Federated Learning →