Four papers to add to your reading list from AI researchers at Meta at #ICLR2024. • ICLR Outstanding Paper Award: Vision Transformers Need Registers ➡️ https://go.fb.me/6kqyj6 • ICLR Outstanding Paper honorable mention: Flow Matching on General Geometries ➡️ https://go.fb.me/itmlgi • Demystifying CLIP Data ➡️ https://go.fb.me/ht0130 • Revisiting Feature Prediction for Learning Visual Representations from Video ➡️ https://go.fb.me/20jah3
Excited to dive into these papers from AI at Meta at #ICLR2024! 📚 As a software development company specializing in AI, Abstrabit Technologies is always eager to learn and innovate. Thanks for sharing these insights! 🚀
Great paper. Maybe it could support one of the possible justification of why CoT-like techniques work in LLM: it could be that the generation of additional tokens is needed by the LLM to create registry tokens to use them to solve the input problem (yet supported by other works). Francesco Corallo
isn't state of a state space model is handled by a register for better optimisation in terms of transformers?
💎 There is something useful and worth paying attention to.
Seems similar to rental saturation and maybe pre-processing? Still getting there, but it makes sense. Good work, team!
Thanks for sharing these insightful papers from Meta's AI researchers at #ICLR2024. Excited to add them to my reading list!
Ummm that Riemann Flow Matching is a powerful mathematical tool! This reminds me of loop integrals and non-normalizability in QFT…seems like a great application for this method!
Thanks for sharing. Vision transformers are really informative read Raktim P.
PhD Student @ PoliTO | Ex Meta
1moVery cool! Let's also check out my AI at Meta internship's paper "Jointly Training Large Autoregressive Multimodal Models" https://arxiv.org/abs/2309.15564 (Friday Morning Poster Session)