The scenario describes the core mechanism of deep learning models: progressive transformation of data through multiple internal stages to extract increasingly abstract features. This functionality is specifically performed by the hidden layers of a neural network.
In a typical deep learning pipeline:
The input layer receives raw or preprocessed data in numerical form but does not perform complex transformations The hidden layers perform a series of mathematical operations (such as weighted sums and activation functions) that transform the data into higher-level feature representations The output layer produces the final prediction or classification result The key phrase in the question is "intermediate representations are repeatedly transformed" and "capturing increasingly abstract patterns." This directly corresponds to hidden layers, which are responsible for feature extraction and hierarchical learning.
As data flows through successive hidden layers, the model learns:
Low-level features in early layers
More complex patterns in deeper layers
High-level abstractions closer to the output
This layered transformation enables deep learning models to handle complex, non-linear relationships in data, such as image recognition, natural language understanding, and predictive analytics.
Therefore, the correct answer is Hidden layers, as they are the components responsible for progressive internal transformation and abstraction in deep learning models.
=========