What Are Learned Representations in AI?
A representation is how a system encodes information so later steps can use it. Representation learning means useful encodings are discovered from data — not only chosen by hand.
Feature engineering vs representation learning
Feature engineering is when people decide which measurements matter. Representation learning is when the model adjusts its own encoding during training, capturing patterns that are hard to specify in advance.
How neural networks learn features
As weights change, the network’s internal features shift toward what helps the task. What information matters can change when the goal changes — the same raw input, different useful encodings.