I am a Machine Learning Engineer at Meta, working on modern recommendation systems, AI retrieval kernels, ads retrieval, and privacy-aware experimentation. My work combines production machine learning, high-performance inference, statistical testing, and applied optimization.
Previously, I worked as a Research Scientist at JD.com American Technologies Corporation and as a Quantitative Researcher at Alphacrest Capital Management. I received my Ph.D. in Applied Mathematics and Statistics from Stony Brook University.
Current Focus
- LLM-based multimodal retrieval for content recommendation and ads retrieval.
- Sequential recommendation, graph neural networks, multimodal content understanding, ANN retrieval, and high-performance inference.
- Privacy-aware online A/B testing, permutation tests, cumulative metrics, and sensitivity improvements for experimentation.
- Information geometry, stochastic bandits, filters, state space models, and convex optimization.
Experience Highlights
Meta, Machine Learning Engineer
Modern recommendation systems and Ads Core ML, 2022-present.
- Initiated LLM-based multimodal retrieval using Llama 4 fine-tuning with production and human-labeled data.
- Delivered retrieval models and infrastructure for ads and content recommendation, including HSTU-style sequential models, GNNs, multimodal content understanding, online ANN, multi-card multi-ANN, and centroid KNN with reranking.
- Designed adaptive experiments for inference performance tuning and helped reduce large-scale GPU usage.
- Built privacy-aware experimentation methods with aggregation, anonymization, permutation tests, cumulative metrics, and total-window sensitivity.
JD.com American Technologies Corporation, Research Scientist
Real-time recommendation and federated learning systems, 2019-2022.
- Built multi-source recall, two-tower pre-ranking, xDeepFM, MMOE, and Hellinger-UCB bandit methods for real-time recommendation.
- Led major algorithm and infrastructure work for an open-source federated learning platform, including vertical federated random forest, vertical federated neural networks, secure multi-party computation, homomorphic encryption optimization, and secure inference for deep learning layers.
- Supported production recommendation, digital marketing, risk control, telephone debt collection, and smart city applications.
Alphacrest Capital Management, Quantitative Researcher
Systematic trading signal research and portfolio optimization, 2015-2018.
- Built the Triangular Input Balanced predictive filter for daily and intraday US equity prediction.
- Designed random forest and Hellinger-UCB reinforcement learning methods for event-data prediction and execution cost reduction.
- Designed fast mean-variance portfolio optimization algorithms with transaction cost models, multi-core parallelism, and C acceleration.
Selected Writing
- Deep Learning with State Space Model (2) - Literature Review with OpenAI
- Deep Learning with State Space Model (1) - Literature Review with DeepSeek
- RLS Filter by DeepSeek
- Abtest and Information Geometry
- Joining meta and Koopman operator theory
Publications
- Xue, Bi, et al. SilverTorch: A Unified Model-based System to Democratize Large-Scale Recommendation on GPUs. Submitted to VLDB 2026.
- Yang, R., Wang, J., Mullhaupt, A. (2024). HELLINGER-UCB: A novel algorithm for stochastic multi-armed bandit problem and cold start problem in recommender system.
- An Efficient and Robust System for Vertically Federated Random Forest
- Fedlearn-Algo: A flexible open-source privacy-preserving machine learning platform
- Wang, J. (2018). Operator Splitting Method and Its Applications in Quantitative Finance. Doctoral dissertation, State University of New York at Stony Brook.
Service
AAAI 2026, AISTATS 2026, KDD 2024, NeurIPS 2024, and Journal of Dynamics and Games.
Education and Skills
Stony Brook University
Ph.D. in Applied Mathematics and Statistics, 2013-2018.
Production code experience across Python, SQL/Hive/PySpark, Java, and C/C++/CUDA.