Binfeng (Bill) Xu

I'm an AI research engineer. Currently, I work at NVIDIA on autonomous agents.

Formerly, I was a researcher at Samsung Research America (SRA) where I built LLM post-training infra and efficient agentic models. During my MS. at NYU, I briefly worked with Alfredo Canziani and Yann LeCun on autonomous driving. Prior at WFU, I was advised by Grey Ballard on efficient tensor decomposition and Paúl Pauca on object detection for drone & satellite images. In my spare time, I enjoy ML competitions on Kaggle and have ranked top 1% globally.

Over the decade, I have been hands-on with a broad range of AI/ML research and applications. I enjoy training large models on massive data and compute, and to build real world products out of ideas.

GitHub  |  LinkedIn  |  Scholar  | 

LLM Post-training. Lately I've been working and researching on the whole spectrum of post-training algorithms and infra: from data synthesis to Knowledge Distillation (eg. SFT with KL), Preference RLHF (eg. DPO), RLVR (eg. GRPO on hard math), Memory efficient Fine-tuning (eg. LoRA), Inference Acceleration(eg. SSM, Flash Attention, Speculative Decoding), Quantization and Compilation (eg. AWQ, llama.cpp), Model Mergeing (eg. slerp), etc. I've also spent tons of time on product and architecting, and have delivered several core projects for Samsung on-deive RAG, Knowledge Graph QA and agentic chatbots.

LLM Reasoning and Machine Intelligence. Generally bipartites into:

• Conceptual learning: Modeling multimodal world patterns with (very) long-context autoregressive transformers. eg. LWM. Recently, Deepseek embarks a successful trail (GRPO) to self-supervise LLM on verified rewards, at a cost of increased test-time compute.
• Generative reasoning: Symbolic Learning models (eg. AlphaGo) shows the strength at (relatively) small action space, which inspires recent LLM research to annotate data with Tree Search sampling (eg. MCTS) and process supervision. I'm currently more interested in GFlowNet in stochastic prediction of wild world space.

Language Agents: Training large models for autonomous actions with optimized reasoning and planning chains.

• ReWOO : Eliminate stacking redundancy in ALM systems by decoupling LLM reasoning from observations.
• Gentopia : A collaborative agentic framework to build hierarchical agents through config, integrating model specialization, evaluation, sharing and inheritance. [demo].

Petting Kobu, the Norway Forest 🐱; Cyberpunk; Digital nomad (someday); Fan of all games by Hidetaka Miyazaki, who motivated me once into indie game devs; Good at Dota2 (once). Photography @500px; Minimalist;