GANs MCQs | JoinMyTutor
MCQs on Generative Adversarial Networks (GANs).
GANs & Advanced Generative Models
MCQs on GANs and advanced generative models.
-
What does GAN stand for?
- A. Generative Adversarial Network
- B. Graph Attention Network
- C. Gradient Approximation Network
- D. Generalized Auto Network
Answer: A — GANs are composed of a generator and a discriminator competing with each other.
-
Which component generates fake samples?
- A. Discriminator
- B. Generator
- C. Encoder
- D. Decoder
Answer: B — The generator creates synthetic data to fool the discriminator.
-
Which component classifies real vs fake?
- A. Discriminator
- B. Generator
- C. Encoder
- D. Decoder
Answer: A — The discriminator evaluates the authenticity of input samples.
-
GANs are typically trained using which method?
- A. Supervised Learning
- B. Unsupervised Learning
- C. Adversarial Training
- D. Reinforcement Learning
Answer: C — The generator and discriminator are trained adversarially.
-
Which loss is used for GAN training?
- A. MSE
- B. Cross-Entropy
- C. Hinge
- D. All of the above
Answer: D — Various losses like BCE, MSE, or Hinge can be used depending on the GAN variant.
-
GAN stands for?
- A. Generative Adversarial Network
- B. Gradient Adaptive Network
- C. Graph Attention Network
- D. Generative Attention Network
Answer: A — GANs use generator and discriminator in adversarial setup.
-
Which loss is used in classic GAN?
- A. Mean Squared Error
- B. Binary Cross-Entropy
- C. KL Divergence
- D. IoU Loss
Answer: B — Binary Cross-Entropy is commonly used for discriminator training.
-
StyleGAN is used for?
- A. Text generation
- B. High-quality image synthesis
- C. Reinforcement learning
- D. Time series prediction
Answer: B — StyleGAN generates realistic high-resolution images.
-
Wasserstein GAN improves?
- A. Training stability
- B. Data augmentation
- C. Gradient clipping only
- D. Memory usage
Answer: A — WGAN addresses mode collapse and improves training stability.
-
CycleGAN is used for?
- A. Image-to-image translation without paired data
- B. Text classification
- C. Speech recognition
- D. RL policy
Answer: A — CycleGAN can map images between domains without paired examples.
-
Generative models for audio/video include?
- A. GANs, diffusion models, autoregressive networks
- B. Only CNNs
- C. Only RNNs
- D. Only transformers
Answer: A — Generative models produce synthetic audio/video data using advanced architectures.
-
Diffusion models work by?
- A. Iteratively denoising random noise to produce samples
- B. Pooling layers
- C. Matrix factorization
- D. Random forest
Answer: A — Diffusion models start from noise and gradually produce realistic outputs.
-
Applications include?
- A. Music generation, video synthesis, deepfakes
- B. Only classification
- C. Only segmentation
- D. Only OCR
Answer: A — Generative models are widely used in creative content generation.
-
GANs consist of?
- A. Generator and discriminator networks
- B. Encoder only
- C. Decoder only
- D. Single classifier
Answer: A — GANs use two networks in a minimax game to generate realistic samples.
-
Challenges include?
- A. Mode collapse, high compute, quality assessment
- B. No training
- C. Only small data
- D. No evaluation
Answer: A — Generative modeling is computationally intensive and may face instability.
💡 Tips & Common Mistakes
Remember to review concepts carefully. Fill in custom tips for gans here.
📅 Last Updated: November 20, 2026
Vinod Kumar
Senior Full-Stack InstructorPython & JavaScript Tutor at JoinMyTutor
Vinod teaches practical Python and JavaScript—ranging from basics to advanced web apps—with a focus on clear mental models, code readability, and interview preparation. He authors both the Python and JavaScript tracks on JoinMyTutor.