2023-05-23-ResearchSeminar-Wellcome Sanger Institute
in Talks
An invited research sharing at Wellcome Sanger Institute.
Multi-domain alignment, LLMs, and generative AI (text-to-image).
Key advances in multi-domain alignment, LLMs, and generative AI:
- Conditioning dropout, as a classifier-free guidance/conditioning.
- Guidance weight: to control the strength of guidance/alignment.
Core topics:
- Diffusion model,
- CLIP model,
- DALL·E 2 (unCLIP),
- Imagen,
- VAE,
- LLMs,
- Multi-domain alignment.
