Akshita Gupta

Career update: PhD thesis submitted, and I'm joining Google as a Research Scientist in @fedassa's team, working on 3D Scene Understanding for the next generation of AR and VR applications.

@CSProfKGD Can meet over coffee around 1, if you have some time.

@akshitac8 Arriving at around 11 this morning for a couple of hours.

Visiting Google Zurich today. My last visit, in 2018, I attended a public talk on something new called the Transformer. You could say it got everyone’s attention.

[1/5]🧵 🔍 Normalization layers have become ubiquitous in modern deep learning architectures, but do they always improve training? In our #ICML2026 paper, we show that, in sparse neural network training, Normalization Layers can actually hurt optimization and slow down convergence. If you're attending ICML, stop by our poster! 📍 Poster #4105 🗓️ Tue, Jul 7, 2026 🕑 2:00–3:45 PM KST 📍 Hall A

Join us tomorrow! 🚀

Great opportunity to work on some fundamental research with a great team!

📢 Call for Papers: WiCV @ #ECCV2026 We welcome: - Extended abstracts - Full-length papers ⌛️ Submission Deadline: July 23 (THU), 2026 🔗 Learn more and submit: sites.google.com/view/wicv-eccv…

Join us for WICV @eccvconf 🎉🎇

You know your former students like you when they reach out to meet up ❤️ Thanks @imAArora (former MSc student) and @akshitac8 for a FUN morning in Zurich!

🎉 Meet our WiCV 2026 at @CVPR speakers! We are honoured to welcome Cordelia Schmid, Sarah Parisot, Georgia Gkioxari, Sanja Fidler, and Katie Bouman for inspiring talks and discussions on the future of computer vision. sites.google.com/view/wicv-cvpr… See you at CVPR! #WiCV #CV

If you want to work with our group while living in New-York and also spend time in Tokyo and Germany!, check out the new positions available at the NYU and Flatiron Institute by @cosmo_shirley post-doc: apply.interfolio.com/184211 research-scientist: apply.interfolio.com/184213

Really excited that our workshop on weight-space symmetries @weightsymmetry is one of the accepted workshops this year at @icmlconf. Please consider submitting your work, see you all in Seoul!

I will be presenting "Generative Re-Photography with Video Models" at several places over the next month. Hit me up if you are around! MIT Media Lab-Apr 1 Harvard-Apr 2 MIT CSAIL-Apr 7 Cornell Tech-Apr 9 Princeton-Apr 10 CMU-Apr 13 Stanford-Apr 24 Berkeley-Apr 28

@CSProfKGD @imAArora @ELLISforEurope @TUDarmstadt It was really nice catching up 😍 see you again in Zurich 😬.

Got a chance to catchup with my former MSc student @imAArora and @akshitac8 at #FOMO2026 in Amsterdam ❤️ Now both @ELLISforEurope PhD students in @TUDarmstadt 💪

We're launching the European Student Robotics Association (ESRA, @esra_robotics) (13 universities, 8 countries, 2.5k+ members)! We’re young, driven and together we’re tackling the European fragmentation problem heads on. Who are we? - ETH Robotics Club (Zürich) @ethroboticsclub - RoboTUM (Münich) - EPFL AI Team (Lausanne) @epflaiteam - Unaite (Paris) - Team Polar (Eindhoven) - TU Wien Robotics Club (Vienna) - Robotics Collective (Aachen) @robocollectiv - KTH AI Society (Stockholm) @KTHAISociety - Delft Robotics Student Association - KN CybAiR (Poznan) - AEA Polimi (Milan) What do we do? → Pan-European robotics competitions → Cross-border technical project collaborations → Coordinated access to funding opportunities across Europe And this is just the beginning! Thanks @andreasklinger, @lukas_m_ziegler and @IlirAliu_ for helping us spread the word :)

Meet Lisa Koch, Asst. Professor Bern Uni 🇨🇭 & ELLIS Member. Her research: ML for biomedical data & medical imaging, with a focus on trustworthy AI. Her goal: "certifiably safe & effective ML tools aligned with clinicians' & patients' needs." #WomenInELLIS

It’s all about community building. Hanging out with the #FoMo2026 crew. Good times ❤️

Career Update: After 6 formative years at FAIR, I'm joining @sainingxie, @ylecun and @jingli9111 at @amilabs to work on the second paradigm of pre-training: world modeling. Next word prediction gave us language intelligence. Next physical state prediction is how we get physical intelligence.

Say hi to Exclusive Self Attention (XSA), a (nearly) free improvement to Transformers for LM. Observation: for y = attn(q, k, v), yᵢ and vᵢ tend to have a very high cosine similarity Fix: exclude vᵢ from yᵢ via zᵢ = yᵢ - (yᵢᵀvᵢ)vᵢ/‖vᵢ‖² Result: better training/val loss across model sizes; increasing gains as sequence length grows. See more: arxiv.org/abs/2603.09078