Meihua Dang
@meihuadang
Ph.D. student @StanfordAILab | Previous M.S. student @UCLA StarAI Lab
Proposing Ctrl-G, a neurosymbolic framework that enables arbitrary LLMs to follow logical constraints (length control, infilling …) with 100% guarantees. Ctrl-G beats GPT4 on the task of text editing by >30% higher satisfaction rate in human eval. arxiv.org/abs/2406.13892
Excited to share that the checkpoints are online at 🤗 HuggingFace. Feel free to check it out! SDXL-DPO: huggingface.co/mhdang/dpo-sdx… SD1.5-DPO: huggingface.co/mhdang/dpo-sd1…
Excited to announce DPO has gone multi-modal! New paper out on RLHF for text-to-image diffusion models! We obtain large-scale state of the art results with 70% win rates against Stable Diffusion XL on human evals! Deep dive below 🧵
#ICLR2023 "Scaling Pareto-Efficient Decision Making via Offline Multi-Objective RL". Proposing offline agents that handle conflicting objectives and new MORL dataset & benchmark. On Monday, May 1st. MH1-2-3-4 openreview.net/forum?id=Ki4oc… @meihuadang @adityagrover_
Reliable control of large language models is a crucial problem. We propose GeLaTo (Generating Language with Tractable Constraints), a neuro-symbolic framework that allows LLMs to generate texts following logical/lexical constraints with 100% guarantee. See arxiv.org/abs/2304.07438
Join @meihuadang @guyvdb and me at @2020Pgm to chat about structured-decomposable #ProbabilisticCircuits, what they are, which #tractable inference they support and most importantly how to #learn them from data! paper: pgm2020.cs.aau.dk/wp-content/upl… code: github.com/UCLA-StarAI/St… (SOON)