Core concepts and principles, Difference between Discriminative and Generative models, Key challenges and opportunities

Image generation (e.g., DALL-E, Stable Diffusion), Text generation (e.g., GPT, Claude), Music synthesis (e.g., Jukebox, MusicLM)

Comparative strengths and weaknesses, Use cases for each approach, Hybrid architectures

Encoder-Decoder structure, Bottleneck layer and dimensionality reduction, Denoising autoencoders

Latent space representation, KL divergence and the reparameterization trick, Generating new samples

Image reconstruction and super-resolution, Anomaly detection in medical imaging, Dimensionality reduction for visualization

Generator & Discriminator dynamics, Adversarial training framework, Common architectures (DCGAN, StyleGAN)

Loss functions (minimax, Wasserstein), Mode collapse and convergence issues, Training tricks and stabilization techniques

Image synthesis on MNIST/CIFAR, Evaluating GAN outputs (FID, IS), Transitioning to more complex datasets

Masked language modeling for text, Vision transformers (ViT) for images, Denoising diffusion models

SimCLR framework for representation learning, MoCo (Momentum Contrast), Applications in few-shot generation

Multimodal generation (text-to-image, image-to-text), Few-shot and zero-shot generation, Ethical considerations in generative AI