Reference Figure 1.

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

Question

Reference Figure 1.

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

BlackTom AI · Accepted Answer

The statement to evaluate is: 'You know that your GAN is trained when D(G(z)) is close to 1.'
Option 1: True. This would imply the discriminator classifies generated samples as real with high confidence. However, in the standard GAN training dynamics, when the generator is well trained and the game reaches equilibrium, the discriminator cannot reliably distinguish real from fake, and D(G(z)) tends toward 0.5, not 1. Therefore, this option misrepresents the training signal.
Option 2: False. In a trained GAN under the original minimax formulation, the generator aims to minimize log(1 - D(G(z))) while the discriminator tries to maximize log(D(x)) + log(1 - D(G(z))). At convergence, the discriminator is fooled equally by real and generated data, leading to D(G(z)) ≈ 0.5 for generated samples. Consequently, claiming D(G(z)) close to 1 indicates the discriminator still strongly believes generated data are real, which would imply imperfect or unstable training rather than a proper trained state.
Option 3: If D(G(z)) were close to 0, that would mean the discriminator reliably identifies generated data as fake, which is also not a sign of a trained generator. Persistent D(G(z)) near 0.5 is a hallmark of balance, not extreme certainty in either direction.
Option 4: In practice, depending on the exact GAN variant (e.g., non-saturating loss, Wasserstein GAN, etc.), the interpretation of D(G(z)) can differ, but the classical criterion for a well-trained GAN under the original formulation is not D(G(z)) ≈ 1. It is approximately 0.5 for generated samples, reflecting a generator that fools the discriminator about half the time.

CS-7643-O01, OAN, OSZ Quiz #5: Module 4

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

查看解析

登录即可查看完整答案

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. 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?

What are the two networks that make up a GAN?

Why were GANs designed to avoid using Markov chains?

更多留学生实用工具

考试浏览器助手

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