Generative AI Models for Trading

This unit develops skills in representing, training, scaling, and combining generative and reinforcement learning models for investment and asset management tasks. Students will work with multimodal market data (prices, video, speech), explore model families (VAEs, flows, GANs, LLMs), and apply optimization and RL techniques to prototype ML-driven portfolio strategies while considering computational and performance trade-offs.

Learning Objectives

• Analyze the conceptual and practical differences between discriminative and generative models and determine appropriate use cases for each in trading and asset-management workflows
• Apply representation learning, dimensionality-reduction, and latent-variable modeling techniques to prepare multimodal financial datasets (market prices, video, speech-to-text) for training generative and reinforcement learning models
• Demonstrate methods for training, scaling, and combining generative architectures (VAEs, flows, GANs, and LLMs), including strategies to optimize model performance and computational cost (e.g., accelerating FinBERT) for financial tasks
• Evaluate and design ML-based portfolio solutions by searching for optimization algorithms, implementing deep reinforcement learning approaches, and assessing LLMs as financial advisors with attention to performance, risk, and regulatory considerations