Which property MOST explains why deep networks can represent complex functions efficiently?

The correct answer is B. Hierarchical feature abstraction. Deep networks build representations layer by layer: early layers detect low-level patterns (edges, tones), middle layers combine them into parts (shapes, textures), and deeper layers assemble those into high-level concepts (objects, faces). This hierarchical composition means a…

NVIDIA Certified Associate (NCA) Core AI Concepts

Question

Options

AUniversal approximation theorem
BHierarchical feature abstraction
CRandom initialization
DHigh numerical precision

How the community answered

(53 responses)

A
2% (1)
B
87% (46)
C
8% (4)
D
4% (2)

Explanation

Deep networks build representations layer by layer: early layers detect low-level patterns (edges, tones), middle layers combine them into parts (shapes, textures), and deeper layers assemble those into high-level concepts (objects, faces). This hierarchical composition means a deep network can represent exponentially complex functions using a manageable number of parameters. The Universal Approximation Theorem (A) guarantees that even a single wide hidden layer can approximate any function, but it says nothing about efficiency - it may require an impractically huge number of neurons. Random initialization (C) is a training technique, not an expressive-power property. High numerical precision (D) is a hardware concern unrelated to representational capacity.

Topics

#Deep Learning#Neural Network Architecture#Feature Learning#Representational Power

Community Discussion

No community discussion yet for this question.

Full NCA-AIIO Practice