Essentials of GenAI

What is the primary goal of Generative AI?

• Classifying data into categories

• Generating new data similar to training data

• Reducing computational costs

• Optimizing reinforcement learning

Which of these is a generative model?

• SVM (Support Vector Machine)

• Random Forest

• GAN (Generative Adversarial Network)

• Logistic Regression

In a GAN, the Generator’s role is to

• Discriminate between real and fake data

• Generate synthetic data to fool the Discriminator

• Compress input data

• Classify images

The Discriminator in a GAN is responsible for

• Creating fake samples

• Distinguishing real data from generated data

• Reducing data dimensions

• Training the Generator directly

Autoencoders are mainly used for:

• Supervised learning

• Unsupervised representation learning

• Reinforcement learning

• Time-series prediction

Which part of an Autoencoder reduces input dimensionality?

• Decoder

• Encoder

• Latent space

• Loss function

How does a VAE differ from a standard Autoencoder?

• VAEs have no decoder

• VAEs use probabilistic latent variables

• VAEs are only for classification

• VAEs don’t use neural networks

The latent space in a VAE is

• Fixed and non-learnable

• Deterministic

• Probabilistic (e.g., Gaussian distribution)

• Discrete

Which loss function is used in VAEs?

• Only Mean Squared Error (MSE)

• Cross-Entropy Loss

• KL Divergence + Reconstruction Loss

• Hinge Loss

"Mode collapse" in GANs refers to

• The generator producing diverse outputs

• The generator producing limited varieties of samples

• The discriminator failing completely

• The model overfitting