Generative Models and Trading Applications

This unit explores generative model families, training and scaling strategies, and their application to trading and asset management using multimodal financial data. Students will learn representation and complexity reduction, latent-variable approaches, GANs and flows, deep reinforcement learning, and practical techniques for accelerating and integrating LLMs (e.g., FinBERT) into portfolio optimization and advisory workflows.

Learning Objectives

• Analyze differences between discriminative and generative models and articulate their implications for forecasting, anomaly detection, and synthetic-data generation in financial contexts
• Apply representation learning, dimensionality-reduction, and latent-variable modeling techniques to preprocess and format multimodal financial datasets (time series, video, and speech) for model training and evaluation
• Demonstrate training, optimization, and scaling strategies—including linear flows, GAN theoretical insights, deep reinforcement learning methods, and techniques to accelerate LLMs—for building generative models tailored to asset-management tasks
• Evaluate and integrate multiple model variants (LLMs, FinBERT, GANs, flows, and deep RL agents) to design, test, and recommend ML-driven portfolio optimization and LLM-based financial advisory solutions while considering performance, computational cost, and practical deployment constraints