Denoising Diffusion Probabilistic Models.pdf

Whether you're a researcher, practitioner, or enthusiast eager to delve deeper into the mechanisms underlying diffusion processes, this workshop offers a comprehensive exploration. This workshop will feature a blend of lectures, interactive discussions, and practical exercises, led by beloved GDSC AI/ML team. Whether you're a novice or an experienced practitioner, this workshop promises to deepen your understanding of diffusion models and inspire new avenues for research and application. Join us and unlock the secrets of diffusion dynamics!

1. Denoising Diffusion
Probabilistic Models

AI and ML Team
Shriram (Head)
Samyuktaa
Sanjai Balajee
Shreyas Sai
Sushmithaa P

2. Image in Math Terms

3. Probability Distribution
Some of the Common Distributions:
→ Uniform
→ Bernoulli
→ Binomial
→ Normal
→ Poisson
→ Chi Squared
Associated Terms: PMF, PDF, CDF

4. Image as a distribution

5. Transport Maps
Types of Modelling

6. Generative Modelling

7. Transfer Modelling

8. Optimal Transport Problem

9. What can we learn from
Thermodynamics

10. Standard Brownian Motion

11. Langevin Dynamics

12. Fokker Plank Equation

13. Noising and the Forward Process

14. Denoising and the Reverse Process

15. Connection with Stochastic Gradient Langevin Dynamics

16. Parametrization

17. Why are Diffusion models better than
GANS?
●Diffusion Models offer fine-grained control over the generation process
●GANs may suffer from mode collapse, where the generator produces
limited or repetitive samples
●Currently they are still slower than GANs at sampling time due to the use
of multiple denoising steps

18. Diffusion Tools
●DALL-E 3: Developed by OpenAI
●Stable Diffusion: Developed by Stability AI

19. ●Graphic design
●Film and animation
●Music and sound design
●Media and gaming industry
Use Cases

20. → Flow matching for generative modelling, Lipman et.al

→ Flow straight and fast: learning to generate and transfer data with rectified
flow. Liu et. al


Give it a read