Tiêu đề M4 - Generative Models - Intro.pdf ✅

What is a Generative Model? • Generative Artificial Intelligence (AI) is a branch of AI technology that focuses on creating something new, such as images, text, and music. • A generative model in machine learning is like an artist that studies patterns in data and then learns to create new, synthetic data that resembles the original. • Unlike discriminative models—which focus on making predictions or classifications— generative models aim to understand the underlying data distribution and generate new samples from it. Example: Real-life Analogies If you gave a generative model thousands of landscape photos, it could learn to create brand- new ones that look realistic—even though they’ve never existed before. It’s like teaching someone the style of Shakespeare and asking them to write a “new” play in that voice. What It Does • Learns probability distribution: It tries to model how data points are actually distributed. • Generates new data: Once trained, it can produce new examples that look like they came from the original dataset. Use Cases • Image synthesis (fake faces, artwork) • Text generation (chatbots, stories, poems) • Data augmentation (generate extra training data) • Drug discovery (generate molecular structures) Popular Generative Models Model Type Description GAN (Generative Adversarial Network) Two neural nets—one generates, one critiques. They battle to improve realism. VAE (Variational Autoencoder) Learns compressed representations and decodes them into new samples. Autoregressive Models (like GPT) Predicts the next element in a sequence, great for text and time series. Diffusion Models Slowly degrade data then learn to reconstruct it— used in image generation.