Ep. 11: Principals of Diffusion Models

About this episode

This episode provides a comprehensive overview of diffusion models and their foundational principles through various perspectives.

This episode provides a comprehensive monograph on diffusion models, detailing their foundational principles through three unifying perspectives: the Variational View (related to VAEs and DDPMs), the Score-Based View (rooted in EBMs and Score SDEs), and the Flow-Based View (connecting to Normalizing Flows and Flow Matching). The core concept involves defining a continuous forward process that adds noise and then learning a corresponding reverse process—a Stochastic Differential Equation (SDE) or Probability Flow Ordinary Differential Equation (PF-ODE) —to transform noise back into data. Much of the discussion focuses on the mathematical equivalence of these different formulations, the tractable training objectives (like Denoising Score Matching), and advanced techniques for accelerating the slow sampling process, including sophisticated numerical ODE solvers (like DPM-Solver) and distillation methods (such as Consistency Models). Finally, the monograph explores the theoretical connection between diffusion models and Optimal Transport (OT) , suggesting that diffusion is related to, but not generally equivalent to, solving the optimal transport problem. Reference: Lai, C. H., Song…

Topics covered

• diffusion models
• Variational View
• Score-Based View
• Flow-Based View
• Stochastic Differential Equation
• Probability Flow Ordinary Differential Equation
• Optimal Transport

Keywords

• mathematical equivalence
• tractable training objectives
• Denoising Score Matching
• sampling process
• numerical ODE solvers
• distillation methods
• Consistency Models

Mentioned in this episode

Products: DPM-Solver

Books & works: The Principles of Diffusion Models

More episodes of AI Extreme Weather and Climate

• Target Concept Tuning: Solving the AI Blindspot in Extreme Weather Forecasting · March 24, 2026 · 0 min
• NeuralGCM: Observation-Based Hybrid Modeling for Global Precipitation Forecasting · January 15, 2026 · 15 min
• Flow-Matched Neural Operators for Continuous PDE Dynamics · December 9, 2025 · 12 min
• Ep 10. RainSeer: Physics-Guided Fine-Grained Rainfall Reconstruction · October 9, 2025 · 19 min
• Ep. 9: FlowCast-ODE Cntinuous Hourly Weather Forecasting with Dynamic Flow Matching and ODE Integration · September 21, 2025 · 16 min
• Ep.8 AQUAH: An Automatic Quantification and Unified Agent in Hydrology · September 3, 2025 · 16 min