Surbhi Mittal

@kmr_dilip India’s AI ambition needs cultural rewiring: idolize creators, not curators. Tech leadership isn’t about hashtags or hype, it’s forged in classrooms, code repositories, and quiet labs. Celebrate grit, not glamour. Let’s root for obsessives, not influencers.

Last week, I had the incredible opportunity to present my work at #ECCV 2024 in #Milan, filled with innovative research in computer vision. Attending #DC was particularly enriching; I was mentored by @orussakovsky, whose insight and outlook on my research was invaluable!

Our work on responsible ML datasets has made it to the cover of #Nature Machine Intelligence! This is an important research problem, and being featured on the cover emphasizes its significance. Super proud of Surbhi Mittal for conceptualizing the brilliant cover design that perfectly represents our work! Link to the journal: nature.com/natmachintell/… Link to our paper: nature.com/articles/s4225… with @surbhimittal97, @thakralkartik78, @RichaSingh26, Tamar Glaser, @cristiancanton, @THassner #MachineLearning #ResponsibleAI #Research #ProudMoment #deeplearning #artificialintelligence

Our latest research, "On responsible machine learning datasets emphasizing fairness, privacy, and regulatory norms with examples in biometrics and healthcare," has been published in Nature Machine Intelligence (Impact Factor 18.8)! We conducted a large-scale audit of biometric and healthcare datasets, spotlighting the critical need for improved curation practices that prioritize fairness, privacy, and regulatory compliance. We've also introduced a responsible rubric for evaluating data quality, providing essential insights for the ethical advancement of AI technologies. "Responsible AI starts with Responsible Datasets" With @surbhimittal97, @thakralkartik78, @RichaSingh26, Tamar Glaser, @cristiancanton, @THassner #AI #MachineLearning #DataEthics #Fairness #Biometrics #Healthcare #AIforGood #iitjodhpur #metaAI Link to the paper: nature.com/articles/s4225…

"On #Responsible #MachineLearning Datasets with #Fairness, #Privacy, and Regulatory Norms": arxiv.org/abs/2310.15848 With Surbhi, @thakralkartik78 , @RichaSingh26, Tamar, @cristiancanton, @THassner #ResponsibleAI #DeepLearning #ResponsibleData #computervision #iitjodhpur #meta

Thank you!🥹 A big shout out to all @WiCVworkshop organizers and participants for creating a collaborative safe space which lets us go beyond our comfort zones …and try new talk formats :)

Throwback to having a great time at the #WICV mentorship dinner at #CVPR2023 with @ilkedemir. Thank you for inspiring us! Your energy is truly unmatched! 💫 #wicv #cvpr2023

very hard to do something novel for the first time; orders of magnitude easier to copy something that is known to be possible. conviction is either extremely difficult, or free.

Gave an invited talk on "Unravelling the Positive Impacts of Computer Vision for Social Good" at 40th foundation of @CDOT_India. I focused on #deepfakes and our face recognition work on #odishatrainaccident. The energy of @devusinh was impeccable. @DoT_India @AshwiniVaishnaw

Self-supervised learning (SSL) is popular recently. We show that an attacker can manipulate (poison) a small part of the unlabeled data with a small patch to inject a backdoor into an SSL model (e.g., MoCo), that a victim will train in the future. #CVPR22 web.cs.ucdavis.edu/~hpirsiav/pape…

We created data2vec, the first general high-performance self-supervised algorithm for speech, vision, and text. When applied to different modalities, it matches or outperforms the best self-supervised algorithms. Read more and get the code: ai.facebook.com/blog/the-first…

Focal length can strongly affect appearance, especially in portraits. This gif ranges 20mm to 200mm focal length. This is why we're sometimes surprised by someone's appearance in person, when we've only seen them in one (or a few) pictures. 2/2 c: reddit /educationalgifs

Bolts is a new Deep Learning research and production toolbox from PyTorch Lightning. Iterate faster with pre-trained models, components, callbacks, and data sets, all modular, tested, and optimized for GPUs/TPUs. Simply subclass, override, and train. medium.com/pytorch/pytorc…

🔥 Introducing... the ML Reproducibility Challenge 2020! The 4th annual edition now expands to cover papers from 7 major ML conferences: NeurIPS, EMNLP, ACL, ICML, ICLR, CVPR and ECCV. Find more here at Papers With Code: paperswithcode.com/rc2020

A relatively large list of computer vision pre-trained models, categorized by deep learning framework (TensorFlow, Keras, PyTorch, Caffe, MXNet). Maintained by Bala Venkatesh (@balavenkatesh22). github.com/balavenkatesh3…

The Phase 1 of the @moderna_tx #SARSCoV2 vaccine results published @NEJM today and they are very encouraging for immune response and safety in 45 healthy people, 2 shots, increasing doses @KPWaResearch nejm.org/doi/full/10.10…

Our new paper (w/@kdexd) argues that "language is all you need" for good visual features: we train CNN+Transformer *from scratch* on ~100k images+captions from COCO, transfer the CNN to 6 downstream vision tasks, and match/exceed ImageNet features despite using 10x fewer images!

@hughlaurie Happy birthday, Hugh! ❤️ #HappyBirthdayHugh #HouseMD

For every hour we spend agonising over societal crimes of the past, we should probably allow a minute or two to wonder what we're doing now that will be similarly condemned a hundred years hence. Climate, obviously; animal rights, likewise. What else?