Saquib Sarfraz

This paper argues research should pause developing new Low-Rank Adapter merging methods and instead understand *when* reusing Low-Rank Adapters works. It shows reusing adapters is ineffective for unseen tasks unless target knowledge or templates are present. Methods 🔧: →…

📢We show that continuous latent reasoning has a theoretical advantage over discrete token reasoning (arxiv.org/abs/2505.12514): For a graph with n vertices and graph diameter D, a two-layer transformer with D steps of continuous CoTs can solve the directed graph reachability…

Happy to share our paper was accepted at CVPRw Computer Vision the Wild 🎉

🚀Excited to share our latest work: ThinkGeo, a comprehensive benchmark for evaluating tool-augmented LLM agents on real-world remote sensing tasks! Project lnkd.in/duNbRMwK Dataset lnkd.in/d_XwCWZ4 GitHub lnkd.in/d5RxRXAS Paper lnkd.in/duitgvEZ 1/n

Happy to share my first 1st author paper In this paper, we explore that Vision-Language Models (Clip/Siglip) fail in understanding basic Image Transforms. [1/4]

A fairer option might have been to remove the culpable author from the accepted papers. This would punish the right person while satisfying the egos of those who have nothing to lose. Sad #CVPR2025

As promised, #CVPR2025 has delivered! 19 papers desk rejected because one of the auhtor irresponsibly reviewed others' work. Kudos to CVPR for punishing the rest of honest authors and their hard work. Congrats to the community for redefining fairness! 🎉

📣 GenMol has arrived. Learn how GenMol speeds up large-scale #drugdiscovery, thanks to its parallel decoding and fragment remasking strategy. Compare it side-by-side with SAFE-GPT in our blog. #NVIDIANIM #NVIDIAhealthcare Deep dive in our blog ➡️ nvda.ws/4hgTrnX

How to deploy an LLM on an edge device, especially in a car? The Mercedes-Benz research team shows us how! They optimized a small language model (SLM) for in-vehicle function-calling: ⭐️Key Highlights: * Model Compression: * Pruned 8 most similar layers (depth-wise pruning)…

Yesterday the hyped Genesis simulator released. But it's up to 10x slower than existing GPU sims, not 10-80x faster or 430,000x faster than realtime since they benchmark mostly static environments blog post with corrected open source benchmarks & details: stoneztao.substack.com/p/the-new-hype…

Please read our statement on the remarks made by Dr. Rosalind Picard at her NeurIPS 2024 invited talk and our commitment to respect, inclusivity, and upholding our values: neurips.cc/Conferences/20…

I was at her talk and was really taken a back when i saw this slide. She asked for all the push back she is getting. It’s great that Rosalind is going through RLHF right now. Hopefully she will be aligned to anti-racism values and ready to give safe talks real soon. #NeurIPS2024

Many people are in the middle of the @CVPR deadline. So I'm sharing my guide to writing a CVPR paper (or any paper). My students have had this for years but I haven't shared it publicly before. I hope you find it useful and write a great paper. #CVPR2025 medium.com/@black_51980/w…

Some people have lost their titles or jobs due to plagiarism, e.g., Harvard's former president. But after this #NobelPrizeinPhysics2024, how can advisors now continue to tell their students that they should avoid plagiarism at all costs? Of course, it is well known that…

Excited to announce the CALL FOR PAPERS for our Special Issue (SI) on ``Visual Domain Generalization in Real-World Applications'' at IJCV. link.springer.com/journal/11263/… Guest editors: Biplab Banerjee Tatiana Tommasi Vineeth N Balasubramanian Kaiyang Zhou Karteek Alahari

65 doctors, nurses, and paramedics told the New York Times what they saw in Gaza. What they saw was a pattern of children being shot in the head.