I had fun writing this blog post with Misha Khodak. We view the 'diverse tasks' problem as an important yet understudied direction in AutoML, and hope that our benchmarking (nb360.ml.cmu.edu) and competition (cs.cmu.edu/~automl-decath…) efforts spur more research activity.

A little over a week ago, we launched the AutoML Decathlon #NeurIPS2022 competition—a competition to develop efficient AutoML methods that work on diverse machine learning tasks for a chance to win a $15,000 top prize! [1/n]

Can we use LLMs to tackle genomics tasks? Vision transformers to solve PDEs? In new work led by Junhong Shen, we consider the general problem of cross-modal fine-tuning and provide surprisingly optimistic answers to these questions. 1/N arxiv.org/abs/2302.05738

#ICML2023 See our poster, *Speeding up Fourier Neural Operators via Mixed Precision* at the SynSML Wksp at 11am on Fri (Room 320)! We improve runtime/memory for FNO by up to 35% 🚀 Renbo Tu Jean Kossaifi Gennady Kamyar Azizzadenesheli Prof. Anima Anandkumar

Ilya Sutskever People building AGI unable to predict consequences of their actions 3 days in advance.

Super excited to share our paper on efficient training of Neural Operators with mixed precision, recently accepted at #ICLR2024! We show more than 50% gain in throughput with almost no loss in accuracy. Neural operators are #AI methods for solving #PDE . Unlike traditional

Checkout our new #ICLR paper on efficient mixed precision training of neural operators. Our method is available out-of-the-box to try with PyTorch in the NeuralOperator library. neuraloperator.github.io/neuraloperator…

Choosing the right precisions in neural operators, We show that, in operator learning, proper reduction of precision improves speed, memory, and often performance, rooted in learning theory and approximation theory, at ICLR 2026 .

I applied to 100 jobs using a resume with the name, "Kismma D. Nhuhts" and I got 29 interviews. This is what I've learned about resumes:

To celebrate William Shakespeare's 460th birthday, here are his 46 best (and strangest) insults:

Come check out our paper on mixed precision neural operators at #ICLR2024! We show >50% increase in throughput with almost no loss in accuracy! Poster # 58, Fri 4:30 - 6:30pm CEST arxiv.org/abs/2307.15034 Renbo Tu Jean Kossaifi Boris, Gennady, Kamyar Azizzadenesheli Prof. Anima Anandkumar

Check out our #ICLR2024 paper on mixed precision neural operators. We show >50% increase in throughput with almost no loss in accuracy! Poster # 58, Fri 4:30 - 6:30pm CEST arxiv.org/abs/2307.15034 Colin White Renbo Tu Jean Kossaifi Boris, Gennady, Kamyar Azizzadenesheli Ours is a

So many new LLM architectures (Mambas🐍, Transformers🤖,🦙,🦔, Hyenas🐺,🦓…), so little GPU time to combine them into hybrid LLMs… Good news! Today we release Manticore, a system for creating **pretrained hybrids** from pretrained models! 👨‍🌾🦁🦂 arxiv.org/pdf/2406.00894 1/n

NEraps 2024

"NeuralDMD"—a fully interpretable neural framework that fuses neural implicit fields with Dynamic Mode Decomposition to recover spatio-temporal dynamics from very sparse and noisy data. From black-hole imaging to weather nowcasting Jw. Ali SaraerToosi, Renbo Tu, and Aviad