Haoyang Fang (方昊扬)

Senior Applied Scientist @ Amazon AGI (formerly AWS AI)

I am a Senior Applied Scientist based in Seattle, working on automating and self-improving how models are trained, with research spanning post-training strategy discovery (SFT and RFT), foundation models, multimodal deep learning, and RAG.

I proposed and led LLMZero, an agentic system that discovers and optimizes adaptive, multi-stage strategies for post-training. Instead of relying on static, expert-tuned recipes, it diagnoses training dynamics, identifies issues, and uses tree search to find better strategy trajectories, improving GRPO by ~17% over the practitioner baseline and generalizing beyond RL to SFT (+4.3%). I am currently extending the approach to broader post-training and foundation-model pretraining.

I also develop frameworks for general end-to-end machine/deep learning automation, such as AutoGluon Assistant (aka MLZero) , a multi-agent system I proposed and led for autonomous end-to-end ML.

Previously, I was a core team member of AutoGluon Multimodal across releases v0.7 to v1.5 .

🔥 News

Sep 2026 New release: Mitra-v2.
Aug 2026 LLMZero accepted by EMNLP 2026 Findings.
Aug 2026 ExTS accepted by EMNLP 2026 Findings.
Aug 2026 Structured Data Foundation Models Tutorial @ KDD 2026.
Jul 2026 MMTS accepted by TMLR.
Jul 2026 OPERA accepted by COLM 2026.
Jun 2026 New preprint: Optimizing CUDA like a Human: Micro-Profiling Tools as Expert Surrogates for LLM-Based GPU Kernel Optimization.
Jun 2026 New preprint: ReSkill: Reconciling Skill Creation with Policy Optimization in Agentic RL.
Apr 2026 AdvEvo-MARL accepted by ICML 2026.
Mar 2026 SenTSR-Bench accepted by AISTATS 2026.
Dec 2025 Re-Table accepted by EACL 2026 Findings.
Nov 2025 Hosting AutoGluon Workshop @ Re:invent 2025.
Nov 2025 Featured in Amazon Science Blog.
Nov 2025 MLZero accepted by NeurIPS 2025.
Nov 2025 Contributed Talk @ AMLC Workshop: 2nd Gen-AI Autonomous Agents.
Nov 2025 Poster Presentation @ Stanford Graph Learning Workshop 2025.
Jul 2024 AutoMM accepted by AutoML 2024.
Jan 2024 Conv-LoRA accepted by ICLR 2024.
Jan 2024 ControlAug accepted by WACV 2024.

📚 Selected Publications

LLMZero: Discovering Adaptive Training Strategies for RL Post-Training via LLM Agents

Haoyang Fang, Wei Zhu, Boran Han, Alex Zhang, Zhenyu Pan, Shuo Yang, et al.

EMNLP 2026 Findings

PDF Code (release soon)

ExTS: Exploit More, Explore Smarter for Budget-Constrained Agentic Search

Haoyang Fang, Bernie Wang

EMNLP 2026 Findings

PDF Code (release soon)

MLZero: A Multi-Agent System for End-to-end Machine Learning Automation

Haoyang Fang, Boran Han, Nick Erickson, Xiyuan Zhang, et al.

NeurIPS 2025

PDF Blog Code

Data augmentation for object detection via controllable diffusion models

Haoyang Fang, Boran Han, Shuai Zhang, Su Zhou, Cuixiong Hu, Wen-Ming Ye

WACV 2024

PDF Code

Convolution Meets LoRA: Parameter Efficient Finetuning for Segment Anything Model

Zihan Zhong, Zhiqiang Tang, Tong He, Haoyang Fang, Chun Yuan

ICLR 2024

PDF Code

OPERA: Online Data Pruning for Efficient Retrieval Model Adaptation

Haoyang Fang, Shuai Zhang, Yifei Ma, Hengyi Wang, Cuixiong Hu, Katrin Kirchhoff, Bernie Wang, George Karypis

COLM 2026

PDF Code

AutoGluon-Multimodal (AutoMM): Supercharging multimodal AutoML with foundation models

Zhiqiang Tang, Haoyang Fang, Su Zhou, Taojiannan Yang, Zihan Zhong, et al.

AutoML 2024

PDF Code

🎓 Education

M.S. in Computer Vision (MSCV)
Carnegie Mellon University (CMU) | 2019 - 2020
B.S. in Pure Math Advanced and Computer Science
Hong Kong University of Science and Technology (HKUST) | 2015 - 2019
Silver Medal, CMO
The Chinese Mathematical Olympiad | Dec 2013

🤝 Community Service

Reviewer: ECCV 2020, ICLR 2022, ICLR 2023, ICLR 2024, ICLR 2025, CVPR 2026, ECCV 2026, COLM 2026, TMLR.

🔥 News

📚 Selected Publications

🎓 Education

🤝 Community Service

🌍 Visitors