While developing a Generative Adversarial Network (GAN) to generate synthetic training images, your generator begins outputting identical or highly similar images repeatedly, failing to capture the full diversity of the training dataset. What is this classic GAN training failure mode called?
Select an answer to reveal the explanation.
Short Explanation and Infographic
If you've ever tried training a Generative Adversarial Network—a GAN—you know it's like trying to balance a marble on a knife-edge. You have two networks, the Generator and the Discriminator, playing a game against each other. If one gets too smart too fast, the whole training process collapses. A classic headache here is "mode collapse." This is when the Generator figures out one single trick—like generating a really good picture of a cat—and just prints that same cat over and over again because it fools the Discriminator, instead of learning to generate dogs, birds, or cars. Keeping the training stable and avoiding this mode collapse is one of the toughest challenges in GAN design.
Full explanation below image
Full Explanation
Generative Adversarial Networks (GANs) consist of two neural networks—the Generator (which creates fake data) and the Discriminator (which evaluates data for authenticity)—trained simultaneously in a zero-sum, adversarial game. This unique setup makes GANs notoriously difficult to train compared to standard feedforward networks.
The two most prominent and interconnected issues in GAN training are: 1. Training Instability: Because the two networks update their weights dynamically in response to each other, they can enter non-convergent cycles where the loss functions oscillate wildly instead of reaching a stable Nash equilibrium. 2. Mode Collapse: This occurs when the Generator discovers a small subset of outputs (a single "mode") that consistently fools the Discriminator. Instead of learning the full diversity of the target distribution (e.g., generating all digits from 0 to 9), the Generator outputs identical or highly similar samples repeatedly (e.g., only generating a specific style of the digit "8").
- Option A is incorrect because while vanishing (disappearing) gradients are a general deep learning problem related to deep networks and activation functions, it is not the specific, defining adversarial failure mode of GANs (though it can occur in GANs, it is mitigated by choosing appropriate loss functions like Wasserstein GAN). - Option B is incorrect because excessive memory consumption is a resource constraint rather than an algorithmic failure mode of the training process itself. - Option C is incorrect because high latency during inference is a hardware or optimization concern, not a training failure mode of GANs.