user avatar
Ekin Dogus Cubuk
Periodic Labs
@ekindogus
Co-Founder of @periodiclabs Past: Lead of materials science and chemistry at @GoogleDeepMind; Google Brain
San Francisco, CA
Joined April 2013
485
Following
9,118
Followers
Posts
  • Pinned
    user avatar
    Ekin Dogus Cubuk
    Periodic Labs
    @ekindogus
    Sep 30, 2025
    I am excited to announce what @LiamFedus and I have been working on: @periodiclabs, a world class team of experimentalists, theorists, and LLM experts. Scientific discovery is inherently an out-of-domain task. Experimental iteration is required for significant advances,
    user avatar
    Liam Fedus
    Periodic Labs
    @LiamFedus
    Sep 30, 2025
    Today, @ekindogus and I are excited to introduce @periodiclabs. Our goal is to create an AI scientist. Science works by conjecturing how the world might be, running experiments, and learning from the results. Intelligence is necessary, but not sufficient. New knowledge is
    78K
  • user avatar
    Ekin Dogus Cubuk
    Periodic Labs
    @ekindogus
    Jun 27, 2019
    Data augmentation is even more crucial for detection. We present AutoAugment for object detection, achieving SOTA on COCO validation set (50.7 mAP). Policy transfers to different models & datasets. Paper: arxiv.org/abs/1906.11172, Code: github.com/tensorflow/tpu…, details in thread.
  • user avatar
    Ekin Dogus Cubuk
    Periodic Labs
    @ekindogus
    Nov 25, 2024
    Interested in working on materials science and AI at Google DeepMind? We’re hiring a research engineer, details here:
    job-boards.greenhouse.io
    DeepMind
    81K
  • user avatar
    Ekin Dogus Cubuk
    Periodic Labs
    @ekindogus
    Nov 29, 2023
    Thrilled to share this work on materials discovery! We found that OOD generalization of GNNs improves predictably, with increasing data from quantum mechanical simulations. These GNNs allowed us to expand the number of known stable materials by an order of magnitude.
    user avatar
    Google DeepMind
    @GoogleDeepMind
    Nov 29, 2023
    Introducing GNoME: an AI tool that helped discover 2.2 million new crystals. 💎 Crystals are found in everything from the chips powering our phones to solar cells creating clean energy. The model also better predicts the stability of new materials. 🧵 dpmd.ai/GNoME-AI
    GIF
    60K
  • user avatar
    Ekin Dogus Cubuk
    Periodic Labs
    @ekindogus
    Nov 8, 2023
    Materials Discovery team at Google DeepMind is hiring. If interested, please apply via the link below:
    job-boards.greenhouse.io
    DeepMind
    109K
  • user avatar
    Ekin Dogus Cubuk
    Periodic Labs
    @ekindogus
    Jul 21, 2021
    New paper on ML & physics at ICML! Learn2Hop: Learned Optimization on Rough Landscapes With Applications to Atomic Structural Optimization We adapt learned optimizers for atomic structural optimization, and compare to baselines from physics. abs: arxiv.org/abs/2107.09661
  • user avatar
    Ekin Dogus Cubuk
    Periodic Labs
    @ekindogus
    Sep 16, 2022
    Finally a task that better IID models do worse on! Not clear why: none of the mechanisms we investigated correlated with perceptual similarity as strongly as just getting a low accuracy on Imagenet (down to ~15% accuracy).
    user avatar
    mechcoder
    @mechcoder
    Sep 16, 2022
    Would you like to learn more about a domain where better ImageNet classifiers transfer worse? Check out our TMLR paper: Do better ImageNet classifiers assess perceptual similarity better? (openreview.net/forum?id=qrGKG…)
  • user avatar
    Ekin Dogus Cubuk
    Periodic Labs
    @ekindogus
    Nov 13, 2025
    We are looking for a condensed matter theorist to join our team at @periodiclabs. Consider applying if you are an expert on applying formal condensed matter theory to real quantum materials.
    53K
  • user avatar
    Ekin Dogus Cubuk
    Periodic Labs
    @ekindogus
    Nov 27, 2023
    UniMat: a diffusion model for generating atomistic systems, led by Sherry (@mengjiao_yang). I'm particularly excited about benchmarking generative models for their utility! In this work, we focus on the ability to discover novel stable crystals.
    00:19
    user avatar
    Sherry Yang
    @sherryyangML
    Nov 27, 2023
    Checkout UniMat -- a unified representation of materials that enables scaling of diffusion models to millions of stable crystal structures. Website: unified-materials.github.io Paper: arxiv.org/abs/2311.09235
    13K
  • user avatar
    Ekin Dogus Cubuk
    Periodic Labs
    @ekindogus
    Apr 16, 2019
    Latest version of the AutoAugment paper is up: arxiv.org/abs/1805.09501. Stop by our oral presentation at CVPR to learn more! Joint work with @barret_zoph @decentralion Vijay Vasudevan and @quocleix.
  • user avatar
    Ekin Dogus Cubuk
    Periodic Labs
    @ekindogus
    May 25, 2023
    Emirhan was in *high school* when we started this project, and he did an incredible job in this work! Thanks to @GoogleAI for the CURe grant. Data augmentation now works on few epochs. Longer training, Tied-Augment can significantly improve mixup/RandAugment (77.6% → 79.6%)
    user avatar
    Emirhan Kurtuluş
    @emirhan_krtls
    May 24, 2023
    Excited to introduce Tied-Augment! A general framework that improves pretraining, finetuning, semi-supervised learning with only a few lines of code, while also enabling data-augmentation to help even few-epoch training. arxiv.org/abs/2305.13520
    20K
  • user avatar
    Ekin Dogus Cubuk
    Periodic Labs
    @ekindogus
    Jun 7, 2019
    Training with additive noise helps with robustness to common corruptions, but hurts clean accuracy. Here we show how to overcome this trade-off, and improve both accuracy and robustness on classification and detection tasks. Led by the outstanding @GoogleAI resident @iraphas13 :
    user avatar
    rapha
    @rapha_gl
    Jun 7, 2019
    2nd @GoogleAI residency paper now on arxiv! "Improving Robustness Without Sacrificing Accuracy with Patch Gaussian Augmentation" 👉🏽 arxiv.org/abs/1906.02611 Patch Gaussian gets SOTA on Common Corruptions & more! Thread for deets 👇🏽
  • user avatar
    Ekin Dogus Cubuk
    Periodic Labs
    @ekindogus
    Apr 13, 2020
    Harvard Üniversitesi'nin bilgisayar bilimlerine giriş dersi (CS 50) artık Türkçe. @kodluyoruz ekibine teşekkürler!
    user avatar
    Kodluyoruz
    @kodluyoruz
    Apr 13, 2020
    Bugün büyük gün: Harvard Üniversitesi’nin efsanevi Bilgisayar Bilimlerine Giriş dersi CS50x artık Türkçe🥳Üstelik yazılıma ilk adımını atmak isteyen herkes için online ve ücretsiz: kodluyoruz.org/cs50! @davidjmalan, gönüllülerimiz ve sponsorumuz @TroyOdeme'ye teşekkürler🚀
    00:00
  • user avatar
    Ekin Dogus Cubuk
    Periodic Labs
    @ekindogus
    Nov 19, 2019
    RandAugment has a significantly smaller search space, which allows it to be optimized on the model and dataset of interest (instead of having to use a smaller proxy task). It works on CIFAR-10/100, SVHN, ImageNet, and COCO.
    user avatar
    Barret Zoph
    @barret_zoph
    Nov 19, 2019
    *New paper* RandAugment: a new data augmentation. Better & simpler than AutoAugment. Main idea is to select transformations at random, and tune their magnitude. It achieves 85.0% top-1 on ImageNet. Paper: arxiv.org/abs/1909.13719 Code: git.io/Jeopl