Yesterday I had the honor of delivering a keynote at the
#NationalAcademies Workshop on Transformative Science & Technology. Workshop organized by tech leaders Bill Dally, Laura McNamara, and Amina Qutub.
I spoke about the evolution of
#generativeAI models —
#VAE →
#GAN →
#DiffusionModels, which evolves from one-step generation to learning a controlled stochastic process for substantially better learnability and expressiveness.
I highlighted our work on showing that diffusion models naturally adapt to and recovers data's low-dimensional manifold structure (
lnkd.in/gY87ZVDf), and how we can add guidance to any pretrained diffusion model to enable design-on-purpose and optimization (
lnkd.in/gb3Vayym).
I also shared examples from our recent research on
#AIforScience and
#AIforEngineering:
- Our Nature Machine Intelligence paper on diffusion language models for decoding genome sequences with Professor
@lecong :
lnkd.in/gQADQDE9 and
lnkd.in/gwkMbqAC
- Our IEEE award-winning paper on diffusion for RFIC design joint with Professor
@KSG_Princeton, enabling AI-native chip design in seconds:
bit.ly/3FoaXt9
It was inspiring to discuss how foundational advances in generative modeling can open amazing new doors, bridging theory, applications, and real-world impact. Excited to continue this journey with the community. 🚀
Workshop link:
nationalacademies.org/event/…