CV

GMorph: Accelerating Multiple DNN Inference via Cross-Task Computation Reuse

Proceedings of the 19th European Conference on Computer Systems

Raccoon: Prompt Extraction Benchmark of LLM-Integrated Applications

Findings of the Association for Computational Linguistics: ACL 2024

Event-Event Relation Extraction using Probabilistic Box Embedding

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)

Unsupervised Mass Product Property Extraction

Develop a unsupervised pipeline that generates formatted properties for billions of listed goods on Pinduoduo & Temu.

• Developed a pipeline that converts property templates into finite state machines to guide LLM's decoding, ensuring legal JSON output for all LLM generated product properties
• Implemented a unsupervised data cleaning process to filter out duplicates and low quality product properties.
• Built a keyword and feature generation process to support billions of products.
• Designed a holistic evaluation pipeline for product properties, including format validation, keyword matching, and semantic similarity.

Red-teaming Benchmark for LLM-Integrated Apps

Analyze prompt injection attacks, develop compound attack strategies, and propose defense templates to enhance LLM safety through in-context learning

• Collect 200+ system prompts and instructions from user created ChatGPT based applications on OpenAI's GPT Store.
• Summarized state-of-the-art prompt injection paradigms into 10+ categories and further constructed compound attacks.
• Derived 20+ prompt templates for in-context defense against prompt injections.
• Built Raccoon benchmark to evaluate safety of LLM-integrated applications across models under complex attack scenarios.

DNN Acceleration via Graph Mutation

Improve DNN inference efficiency for multi-task systems.

• Led the design and implementation of a parser that converts PyTorch models to intermediate graph representations.
• Designed a graph compiler that reuses layers and estimates inference time of new models using sampling.
• Designed an algorithm based on simulated annealing to balance exploration and exploitation of graph merging.
• Utilized frameworks like TensorRT & TVM to further optimize model inference time.

Learning Joint Event Relations with Boxes

Ensure logical consistency across narrative events

• Designed an XML parser and a data loader for extracting labeled relations and improving the overhead of loading data.
• Aggregated data with pandas and visualized principal components with existing labels.
• Designed and implemented experiments that utilizes Longformer to handle larger text input which exceeds RoBERTa's max token size.

RL Algorithms with High Safety Guarantees

Safe ML algorithms with mathematical guarantees.

• Designed and developed machine learning algorithms with high safety constraints using Seldonian Framework.
• Derived concentration inequalities and implemented various Importance Sampling estimators for high confidence policy improvements.
• Developed a library for Seldonian Framework with Numpy and Numba optimizations.
• Opensourced on GitHub with full documentation.

NLP Engineer

Pinduoduo

Design production-level LLM application.

• Implement unsupervised extraction pipeline for mass product properties.
• Construct keyword and feature generation process for billions of products.
• Build end-to-end system for query-product and product-product relevance learning.
• Optimize model inference efficiency via model distillation and quantization.

Software Engineer

Amazon Robotics

Develop next-gen Amazon Grocery Automation System.

• Designed and implemented AWS cloud infrastructure and algorithms to enable ultra fast 1-2 hour grocery delivery to Amazon customers as part of a new automation pilot program.
• Led the design of inventory and order language models.
• Designed an event driven MQTT workflow management system utilizing AWS IoT.
• Led the test infrastructure design to enable end to end testing of order ingestion to automation fulfillment.

Research Intern

Alibaba DAMO Academy

Research RL-based recommendation systems for Alibaba's online second-hand market Xianyu.

• Built the homepage Merchant Feed in Xianyu to a classic Contextual Bandit problem.
• Implemented a recommendation system based on a modified Linear UCB algorithm which utilizes both browsing features and click features.

University of Massachusetts Amherst

Master of Science

Computer Science

University of California, Irvine

Bachelor of Science

Computer Science & Engineering, minor in Statistics

First Prize

Jamming With Ubuntu 2017 in Rugao

Best TIPPERS program

UCI IoT Tippers Hackathon

Best Project and Development Practices

CSULB BeachHacks