Kelly Marshall

SWE @ Google
Formerly NYU Tandon PhD, NYU 上海 undergrad

[Chinmay Hegde @chegday]

Thrilled that Arboretum* was accepted to #NeurIPS2024 (D&B) as a spotlight! Congrats to the whole team.

USDA/NIFA recently highlighted this work here:
https://t.co/AzoIcgdRZF

* Several updates are in the pipeline, including a name change (!!). Stay tuned...

[Kelly Marshall @kellym11111]

@polyphron @KatieConradKS @ArmoredSaint65 Those are all the SOTA methods for erasure. If you're confident you can come up with a better one you're free to try :)

[Kelly Marshall @kellym11111]

@polyphron @KatieConradKS @ArmoredSaint65 I'm afraid it's not that simple, try checking out our latest ICLR work on this: https://t.co/gH5ldNFWK4

[Anushrut Jignasu @ajignasu]

🚨🚨Excited to announce "Towards Foundational AI Models for Additive Manufacturing: Language Models for G-Code Debugging, Manipulation, and Comprehension."

@kellym11111, @baskinscience, Aditya Balu, @chegday , @adarshk

https://t.co/ltMWTkDw16

1/2

[Alex Carlier @alexcarliera]

Gaussian Painters imported into #b3d as ellipsoids

(3D Gaussian Splatting plugin for Blender - work in progress)

[Accepted papers at TMLR @TmlrPub]

Distributionally Robust Classification on a Data Budget

Benjamin Feuer, Ameya Joshi, Minh Pham, Chinmay Hegde.

Action editor: Yarin Gal.

https://t.co/Au0e6vUMne

#robust #robustness #datasets

[Minh Pham @mnphamx1]

Excited to share our latest work with @kellym11111 and @chegday:

“Circumventing Concept Erasure Methods For Text-to-Image Generative Models”

Paper: https://t.co/FKEAQQlRRN

Project website: https://t.co/sPGPkG28kY

Thread 🧵

[Kelly Marshall @kellym11111]

Presenting ZeroForge: A text-to-3D model which doesn’t require training data while avoiding inference-time optimization.

Paper and code below:
https://t.co/xoioMfA125

[Ameya Joshi @ameyaajoshi]

Defended my dissertation today! Thanks everyone at DICE lab and Prof. @chegday.

[Sergey Levine @svlevine]

I wrote a short article about some possibly untapped potential in the combination of RL and language models: https://t.co/WLn8ugzwgr

What if the full power of LMs will come not from their ability to be *generative* per se, but their ability to predict the behavior of people?

[Kelly Marshall @kellym11111]

@togelius That’s hilarious. For some reason all the recent gpt-related news has made me think about that one a lot