Yes, highly multimodal, and that's why Bayesian inference prefers sync data

My first 100 citations

I will share Entropy-MCMC paper on Wed afternoon ICLR 2026. See you then

welcome to Halle B #132

Postdoctoral Opportunity 🚨 We're excited to announce an opening for a Postdoc at the intersection of Machine Learning and Autonomous Driving! This is a unique chance to develop cutting-edge algorithms in deep learning and probabilistic modeling to serve the planning tasks of

Can we be MUCH FASTER and still get well-aligned text in decoding-time alignment? YES, WE CAN! Thrilled to introduce CARDS, a segment-level sampling algorithm to efficiently generate aligned text. 5x faster and 99% win-ties compared to previous methods. arxiv.org/abs/2406.16306

Introducing CARDS, a new method for LLM decoding-time alignment: ✨5x faster in text generation and 99% win-ties in GPT-4/Claude-3 evaluation ✨provably generates high-reward high-likelihood text ✨no retraining/fine-tuning of LLMs 💡Main idea: alignment as a sampling problem

Now we have integrated LOMO!

The codebase for CARDS is available now. You can try this fast and flexible alignment method at: github.com/lblaoke/CARDS

Introducing ETA, an inference-time alignment method for VLM safety: ✨Improve safety without reducing VLM power—reducing unsafe rate by 87.5% in cross-modality attacks and achieving 96.6% win-ties in GPT-4 helpfulness evaluation ✨Induce only a slight increase in inference time

🚀 Thrilled our work is accepted at ICLR2025! We mitigate safety challenges in VLMs with ETA, a plug-and-play inference-time method that minimizes trade-off between safety and general performance (helpfulness and inference time). Big thanks to Bolian Li and Ruqi Zhang !

Excited to see our chapter out! A concise and accessible introduction to Bayes Compute in deep neural networks and deep generative models. Great for statisticians curious about diving in!

Excited to be selected as a Pathways@RSS. I'll also be presenting our latest work, "Energy-Based Transfer for Reinforcement Learning," at the OOD Workshop during the poster session with ID 62 (3:10–4:00 PM today). Read the full paper here: arxiv.org/abs/2506.16590.

Finally, CARDS is accepted by Conference on Language Modeling

Purdue IPAI is hiring postdocs in AI! If you're interested in statistical machine learning or trustworthy AI, and would like to work with me, please get in touch! Applications are due by Sept 1, 2025. purdue.edu/computes/insti…

Welcome to our poster on efficient LLM alignment #COLM2025 #64. My advisor Ruqi Zhang is here.

Excited to give a talk on Oct 14 about Gradient-Based Discrete Sampling! How can we bring the power of Langevin dynamics to discrete spaces? I’ll discuss algorithms like Discrete Langevin and its extensions for multimodal distributions and combinatorial optimization, with

Can we accelerate test-time alignment? YES! 📃paper: Reward-Shifted Speculative Sampling Is An Efficient Test-Time Weak-to-Strong Aligner 🔗arXiv: arxiv.org/abs/2508.15044 📌EMNLP 2025

Excited to be at EMNLP next week! Our work Visco jailbreaks MLLMs using a visual-centric fabricated context, achieving ~90% ASR on strong models. Catch me in Suzhou! 🤔 BTW, I’ll be at the poster session for SSS (P2), since Bolian Li can’t attend.

Continuous diffusion dominates images but fails on discrete data—despite learning continuous gradients that should enable coordinated updates. "CANDI: Hybrid Discrete-Continuous Diffusion Models" explains why and how why hybrid diffusion fixes it! (1/8)