Which statement about Generative Adversarial Networks (GANs) is TRUE?

Question

BlackTom AI · Accepted Answer

Mode collapse refers to lack of diversity in generated samples.

1) Relevant knowledge points and concepts
- GAN architecture: a GAN has two networks, a generator G that produces synthetic samples from random noise and a discriminator D that tries to distinguish real from generated samples. They are trained adversarially (minimax game).
- Training setting: GANs are typically trained on unlabelled data (unsupervised or self-supervised), not on labelled data as a single supervised network would be.
- Common failure modes: mode collapse is a well-known pathology where the generator maps many different inputs to the same or few outputs, producing limited variety. The discriminator does not generate images; it only evaluates them.

2) Step-by-step reasoning
- Identify each option and test it against definitions:
  a) "GANs always underperform when compared to diffusion models" — this is an absolute statement. Performance depends on metrics (fidelity vs diversity), architecture, and task; GANs can produce very high-fidelity samples and are still competitive in many settings, so the absoluteness makes it false.
  b) "GANs consist of a single network trained on labelled data" — incorrect because GANs consist of two networks (generator and discriminator) and usually use unlabelled data.
  c) "The discriminator generates new images" — incorrect: the generator produces images; the discriminator evaluates.
  d) "Mode collapse refers to lack of diversity in generated samples" — matches the standard definition.

3) Calculations / intermediate steps
- No numeric calculations are required. The logical elimination above is the complete intermediate reasoning.

4) Why the correct answer is correct
- Mode collapse is widely documented in GAN literature: the generator learns to produce a small set of outputs that the discriminator cannot reliably distinguish as fake, so diversity collapses. Papers and textbooks define mode collapse precisely as reduced sample diversity.

5) Why other options are incorrect (brief)
- Option 1 is an overgeneralization; relative performance varies. Option 2 misstates GAN structure and training paradigm. Option 3 confuses the roles of generator and discriminator.

Conclusion: The true statement is "Mode collapse refers to lack of diversity in generated samples," which matches the provided correct answer.

Which statement about Generative Adversarial Networks (GANs) is TRUE?单项选择题

类似问题

Reference Figure 1. Typically early in training, the value of D(G(z)) is closer to which of the following values?

Reference Figure 1. You know that your GAN is trained when D(G(z)) is close to 1

Reference Figure 1. Two cost functions are presented in the figure, which one would you use to train your GAN?

Generative adversarial networks (GANs) learn a parameterized density function, p(x).

Why did Elgammal switch from Generative Adversarial Networks (GAN) to Creative Adversarial Networks (CAN) to create variations of the images he collected for AI artwork development?

Which variation of GANs uses two discriminators: one as a typical binary classifier and the other as a score function?

What is the primary objective of the discriminator in a GAN during training?

更多留学生实用工具

智能学习助手

风格化写作助手

论文查重助手

文献引用助手

课堂转译助手

课堂笔记助手

Quiz搜索助手

学校历年真题

智能刷题助手

智能匹配练习

希望你的学习变得更简单