Ligong Han

World models have made impressive progress in video generation, yet they still struggle with a fundamental challenge: memory. In long rollouts, the camera trajectory gradually drifts from the user-specified motion and revisited scenes no longer align with earlier observations. These errors accumulate over time, causing the generated world to steadily lose coherence. 🚀Excited to share our solution MosaicMem 🌍🧠 — our new hybrid spatial memory for video world models. Project Page: mosaicmem.github.io/mosaicmem/ Paper: huggingface.co/papers/2603.17…

What happens when you invite 150 AI economists (Claude Code) to a research conference, give them the exact same data, and ask them to test the same hypotheses? We did just that. The results reveal a new phenomenon: Nonstandard Errors in AI Agents. 🧵👇

How is generative AI reshaping engineering design? In Episode 2 of No Math AI, hosts Dr. Akash Srivastava (@variational_i) and MIT PhD student Isha Puri (@ishapuri101) sit down with Dr. Faez Ahmed (@_faezahmed) from MIT DeCoDE Lab to explore just that. 👇

#ICLR2025 #BayesDL #LLM #ICL Can LLMs enjoy the accuracy of many-shot in-context learning (ICL) with only the inference cost of zero-shot learning. To address this question, we proposed implicit in-context learning (I2CL).

🚀 Big News! Our latest preprint is out: 🧠 “Two Heads Are Better Than One: Test-time Scaling of Multi-agent Collaborative Reasoning” Introducing M1-32B — an LLM fine-tuned for multi-agent collaboration on M500, a dataset of 500 rich reasoning traces. 👇 (1/4)

📣 Excited to share our ICLR 2025 paper "Implicit In-context Learning (I2CL), achieving few-shot performance at zero-shot inference cost! Don’t miss it at poster session Fri 24th 3:00-5:30pm #228! Paper and code is available at: arxiv.org/pdf/2405.14660 #ICLR2025 #LLM #ML

[1/x] 🚀 We're excited to share our latest work on improving inference-time efficiency for LLMs through KV cache quantization---a key step toward making long-context reasoning more scalable and memory-efficient.

[1/x] can we scale small, open LMs to o1 level? Using classical probabilistic inference methods, YES! Joint @MIT_CSAIL / @RedHat AI Innovation Team work introduces a particle filtering approach to scaling inference w/o any training! check out …abilistic-inference-scaling.github.io

If you're interested in Trustworthy LLMs, particularly probabilistic methods, generalization, or calibration, we' d love to see you there! 🤝This is done with @RutgersCS visiting student @Yibin_Wang_ and my PhD student @shihaizhou as well as @ligongh and Dimitris Metaxas

💃 Score-Guided Diffusion for 3D Human Recovery 🕺 Jupyter Notebook 🥳 Thanks to @statho_ ❤ @ligongh ❤ Dimitris Metaxas ❤ 🌐page: statho.github.io/ScoreHMR/ 📄paper: arxiv.org/abs/2403.09623 🧬code: github.com/statho/ScoreHMR 🍊jupyter by modelslab.com: please try it 🐣 github.com/camenduru/Scor…

Score-Guided Human Mesh Recovery (ScoreHMR) is an approach for solving inverse problems for 3D human pose and shape reconstruction. ScoreHMR mimics model fitting approaches, but alignment with the image observation is achieved through score guidance in the latent space of a diffusion model. Paper: ScoreHMR: Score-Guided Diffusion for 3D Human Recovery Link: arxiv.org/abs/2403.09623 Project: statho.github.io/ScoreHMR/ #AI #AI美女 #LLMs #deeplearning #3D

SVDiff: Compact Parameter Space for Diffusion Fine-Tuning Diffusion models are amazing, but customizing them is hard. SVDiff has 2,200x fewer parameters than DreamBooth because we fine-tune with the singular values of the weight matrices arxiv.org/abs/2303.11305 #ICCV2023 (2/3)

📢 Our #SMART101 challenge is now open! 🎉 Join the brightest minds in multimodal reasoning and cognitive models of intelligence to drive AI progress. 🚀 Don't miss out! Challenge closes on Sept. 1. Winning teams will receive prizes! 🏆 eval.ai/web/challenges… #VLAR #ICCV2023 #AI

Can't decide if your representations should be invariant or equivariant to a transformation? Multi-Symmetry Ensembles (MSE) can help you by combining both! ImageNet results below. Humbled to have our work accepted at #ICML2023, @icmlconf. Joint work w/ @MITIBMLab, @MIT and @AWS.

Do you want to estimate the density ratio accurately but are dissatisfied with the results?

@juefeix Thanks, Felix! 🥰

Nice work from friends at Google and Rutgers on SD personalization. 🥰

SVDiff: Compact Parameter Space for Diffusion Fine-Tuning proposed SVDiff method has a significantly smaller model size (1.7MB for StableDiffusion) compared to existing methods (vanilla DreamBooth 3.66GB, Custom Diffusion 73MB), making it more practical for real-world applications abs: arxiv.org/abs/2303.11305

SINE: SINgle Image Editing with Text-to-Image Diffusion Models abs: arxiv.org/abs/2212.04489 project page: zhang-zx.github.io/SINE/

Diffusion Guided Domain Adaptation of Image Generators abs: arxiv.org/abs/2212.04473 project page: styleganfusion.github.io

If you are a PhD student at @MIT (sorry about this constraint) looking for an #internship and are interested in any of the topics listed in this 🧵, please get in touch with me or @HW_HaoWang, as my group is seeking talented students to join us at the @MITIBMLab