Biography

I am currently a PhD candidate in the department of Computer Science, Tulane University. Previously I got my Master’s and Bachelor’s degrees both from Tianjin University. My advisor is Zizhan Zheng. My current research interests include Reinforcement learning, NLP(large language models), Network and transportation optimization, Recommendation systems, Statistics, Deep learning, Optimization, and Operations research. If you have any questions, please feel free to contact me. My email address is txu9@tulane.edu.

Education

PhD candidate, Computer Science, Tulane University, USA (now)

Master of Engineering, Industrial Engineering, Tianjin University, China

Bachelor of Engineering, Industrial Engineering, Tianjin University, China

Research Interests

Reinforcement learning, NLP(large language models), Network and transportation optimization, Recommendation systems, Statistics, Deep learning, Optimization, Operations research

Working Experience

Dalian Neusoft University of Information 
Lecturer in Information Management Department

Internship Experience

Splunk, (USA)
Machine Learning Applied Scientist Intern
Collaborative Open-Source Agents for Enhancing Chain of Thought Reasoning Quality of SPL: Developed a method that LLM open-source agents collaborate to enhance CoT better than GPT4. Specifically, there are N LLM generators, one relative grader for Chain of Thought (COT) grading, and N iterative improvement agents for improving CoT quality. We tried different models like llama3 8B, phi3, gemma2 9B, and Llama3 70B. Our final score shows that we use open-source models to achieve results better than GPT-4o.

MultiPlan, (USA)
AI Research Intern - Applied Large Language Models
Large Language Model Design: According to Meta’s large language model diplomatic game Cicero, designing our own large language model diplomatic game (need to modify input, output, and model structure).
Pricing Recommendation: Design a recommendation system using contextual bandits and the ClaimtoVec model to do pricing recommendations for three phases: Prepay, Postpay, Arbitration
TextoSQL: Fine-tuning a large language model(T5) on our own dataset on problem question(text) to SQL.

China Automotive Technology and Research Center,
Quality Management Engineer Intern

Caterpillar China,
Supply Chain Management Intern

Danfoss China,
IE Intern

Amazon China,
Returned Goods Process Improver

Languages and Skills

Python, R, Pandas, Matlab, C, C++, Java, Tensorflow, Keras, Pytorch, Mysql, Git, OR-tools, Linux, Latex, SPSS, Flexsim, Gurobi, Project, CAD, PROE, Docker, Sklearn, Jira

Publications

Henger Li, Tianyi Xu, Tao Li, Yunian Pan, Quanyan Zhu, Zizhan Zheng. A First Order Meta Stackelberg Method for Robust Federated Learning (Technical Report). ICML 2023 Workshop on the 2nd New Frontiers In Adversarial Machine Learning (ICML-AdvML).

Xu, T., Zhang, D., & Zheng, Z. Online Learning for Adaptive Probing and Scheduling in Dense WLANs. IEEE International Conference on Computer Communications (INFOCOM), 2023.

Xu, T., Zhang, D., Pathak, P. H., & Zheng, Z. Joint AP Probing and Scheduling: A Contextual Bandit Approach. IEEE Military Communications Conference (MILCOM), Nov. 2021.

Xu T, Ozbek O I, Marks S, et al. Spanish-Turkish Low-Resource Machine Translation: Unsupervised Learning vs Round-Tripping[J]. American Journal of Artificial Intelligence, 2020, 4(2): 42-49.

Lin Y, Xu T*, Bian Z. A two-phase heuristic algorithm for the common frequency routing problem with vehicle type choice in the milk run[J]. Mathematical Problems in Engineering, 2015.

Lin Y, Bian Z, Sun S, Xu T. A two-stage simulated annealing algorithm for the many-to-many milk-run routing problem with pipeline inventory cost[J]. Mathematical Problems in Engineering, 2015.

LIN Y, XU T*. Vehicle Scheduling with Time Windows and Inbound Crossing Restrictions in the Milk Run[J]. Industrial Engineering and Management, 2015, 1: 5.