Can you train a generative model using only noisy data? If you can, this would alleviate the issue of training data memorization plaguing certain genAI models.
In exciting work with Giannis Daras and Alex Dimakis we show how to do this for diffusion-based generative models.

Stoked to announce an Agentic Markets workshop Agentic Markets Workshop @ ICML 2024 at #ICML 2024! ICML Conference

📇 Details: sites.google.com/view/amw-2024/…
✏️ Call for papers: due May 17th
📅 Conference: July 26/27th

featuring GOAT speakers like Tuomas Sandholm, Gillian Hadfield, Ian Gemp Constantinos Daskalakis

The call for papers is coming soon… Delighted to confirm the participation of the Greek Prime Minister Kyriakos Mitsotakis and the chair of Greece’s high level AI committee Constantinos Daskalakis at this event. June 20th in Athens 🇬🇷

Athens has a lot of talented AI researchers and developers.

Daron Acemoglu (@MITEcon) will present #PowerandProgress (with Simon Johnson) to the Greek audience at an online event organized IOBE on Wednesday, July 5 5.30pm Athens time Constantinos Daskalakis (@MIT), Nikos Vettas, and I will discuss with Daron #AI , #technology , & #politics

ACM names the latest class of Fellows: congrats to a whole bunch of great theorists, including Boaz Barak, Sandy Irani, Rafael Pass, Samir Khuller, David Mount and especially my advisor Constantinos Daskalakis!!

Also: my awesome colleague Jimmy Lin!
acm.org/media-center/2…

New ICML paper: Score-Guided Intermediate Layer Optimization (SGILO).

We train diffusion models on the latent space of StyleGAN and we show provable mixing of Langevin Dynamics for random generators.

Reconstructions for *extremely sparse* (<1%) measurements.

A thread🧵(1/N)