@JeanKossaifi Profile picture

Jean Kossaifi

@JeanKossaifi

Pushing boundaries of AI & making it available for all @nvidiaAI | ex @Samsung AI, @amazon AI, PhD @imperialcollege | TensorLy creator https://t.co/1epTxX3XFB 👨‍💻

Similar User
Sanja Fidler photo

@FidlerSanja

Shakir Mohamed photo

@shakir_za

Brandon Amos photo

@brandondamos

Dhruv Batra photo

@DhruvBatraDB

Nal photo

@nalkalc

Jakob Foerster photo

@j_foerst

Petar Veličković photo

@PetarV_93

Durk Kingma photo

@dpkingma

Dustin Tran photo

@dustinvtran

Maithra Raghu photo

@maithra_raghu

Dan Roy photo

@roydanroy

Yisong Yue photo

@yisongyue

David Duvenaud photo

@DavidDuvenaud

Arash Vahdat photo

@ArashVahdat

Jakub Tomczak photo

@jmtomczak

Jean Kossaifi Reposted

Our paper on Incremental Fourier Neural Operator is finally accepted at @TmlrOrg! Joint work with @jiawzhao , @ZongyiLiCaltech, @JeanKossaifi and @AnimaAnandkumar Let me summarize the work in a couple of tweets 1/6

Tweet Image 1

Incremental Spatial and Spectral Learning of Neural Operators for Solving Large-Scale PDEs Robert Joseph George, Jiawei Zhao, Jean Kossaifi, Zongyi Li, Anima Anandkumar. Action editor: Jiajun Wu. openreview.net/forum?id=xI6cP… #fourier #learns #turbulent



Jean Kossaifi Reposted

A number of my colleagues @Caltech and I have put together a letter voicing our concern regarding CA SB 1047 (AI Safety Act). Sign the letter and show your support! docs.google.com/forms/d/e/1FAI… We call on @caltech students, postdocs, faculty, staff, and alumni to sign but also leave…


Jean Kossaifi Reposted

Thank you for all the birthday wishes! I had a fabulous birthday with my @Caltech group, with a surprise cake, and discussed how we can make our Neural Operator library even more accessible and capable.

Tweet Image 1
Tweet Image 2

Jean Kossaifi Reposted

It was such a pleasure to do the podcast with @Dr_NeilA on AI+Science. Building up on my @TEDTalks youtu.be/6bl5XZ8kOzI in the podcast youtu.be/Mamxev0IcZM we traced my journey in AI, how I got started on AI+Science, why LLMs today do not understand the physical world, and…


Jean Kossaifi Reposted

Cool work with @sofyadym linktr.ee/sofyadym

Bruno Raffin from @Inria is telling us about making your #AI4Science and numerical simulation pipeline more efficient with online training.

Tweet Image 1


Jean Kossaifi Reposted

Come to our "Equivariant Graph Neural Operator for Modeling 3D Dynamics" poster tomorrow! Please stop by if you are interested!! Wed 24 Jul 1:30 p.m. CEST — 3 p.m. CEST Hall C 4-9 #203 icml.cc/virtual/2024/p…

Tweet Image 1

📣Check out our #ICML24 paper Equivariant Graph Neural Operator (EGNO), a new framework for physical simulation by modeling 3D ODE as temporal functions, w/ zero-shot generalization to any temporal resolution! Paper: arxiv.org/abs/2401.11037 Code: github.com/MinkaiXu/egno 🧵1/n



Jean Kossaifi Reposted

Here is our new 8B Mamba-based Hybrid LLM: Higher MMLU compared to the 8B transformer and long context extension up to 128K sequences. arxiv.org/abs/2406.07887

A 8B-3.5T hybrid SSM model gets better accuracy than an 8B-3.5T transformer trained on the same dataset: * 7% attention, the rest is Mamba2 * MMLU jumps from 50 to 53.6% * Training efficiency is the same * Inference cost is much less arxiv.org/pdf/2406.07887

Tweet Image 1


Jean Kossaifi Reposted

📣Check out our #ICML24 paper Equivariant Graph Neural Operator (EGNO), a new framework for physical simulation by modeling 3D ODE as temporal functions, w/ zero-shot generalization to any temporal resolution! Paper: arxiv.org/abs/2401.11037 Code: github.com/MinkaiXu/egno 🧵1/n


Another great, quality open-source project.

timm 1.0.3 is pushed out the door and almost at 30k🌟 It's been stable for ages, why 1.0 now? I wanted one long promised feature in -- unified feature map extraction (features_only=True) for almost all models, incl ViT 🎉 Accomplished via new API. github.com/huggingface/py…



Jean Kossaifi Reposted

Tensor algebra is the way to scale AI architectures in an efficient way. We have been working on it for more than a decade. Check out tensorly.org led by @JeanKossaifi with @PyTorch framework

Anthropic's Mathematical Framework for Transformer Circuits is all about Kronecker Products; Because they give you a way to model parallel computation (transformer heads). transformer-circuits.pub/2021/framework…

Tweet Image 1


Loading...

Something went wrong.


Something went wrong.