Feiran Jia

I am a Ph.D. student in Informatics at Pennsylvania State University (PSU), working under the guidance of Prof. Anna Squicciarini and Prof. Qingyun Wu. My research spans Computational Game Theory, Social Network Analysis, and Machine Learning, Deep Learning, LLM Agents, and their ethical implications.

Before coming to PSU, I completed my M.S. in Computer Science at Washington University in St. Louis, where I was advised by Prof. Yevgeniy Vorobeychik and Prof. Chien-Ju Ho. I received my B.S. in Computer Science with a minor in Finance from ShanghaiTech University. I also interned at Meta Ads CoreML.

Email  /  CV  /  Google Scholar  /  Github

Research

In Computational Economics, my research applies computational methods to address complex societal challenges, focusing on Game Theory, Social Networks, and their ethical implications. My work in Machine Learning emphasizes efficiency and reliability, particularly in areas like trustworthy AI and AutoML. During the summer of 2024, I worked on enhancing ranking models through uncertainty quantification. More recently, I have been exploring LLM Agents, focusing on their efficiency, behavior, and security.

This paper Investigats human-like trust behaviors and alignment in language model agents through behavioral economics trust games.

The paper proposes RoCourseNet, a system that enhances both prediction and explanation capabilities of machine learning models, ensuring their effectiveness despite data shifts.

The paper introduces a new framework for modeling user decisions in online personal information sharing and proposes welfare-optimizing algorithms.

The project examines test-time adaptation (TTA) and discovers a security vulnerability where malicious data in the same batch can affect predictions on benign samples, resulting in the development of Distribution Invading Attack (DIA).

This research explores using GPT-4 to solve complex math problems, including a newly proposed conversational approach called MathChat.

The project introduces a novel targeted multi-objective hyperparameter optimization approach with lexicographic preference and presents LexiFlow, a randomized directed search method.

This paper develops Structured Hierarchical Games for modeling hierarchical decision-making and introduces a gradient-based method, Differential Backward Induction, to approximate game equilibria effectively.

This paper studies how peer-pressure and random noise affect consensus-building in networked systems.

The paper introduces a multi-level game-theoretic model to analyze hierarchical policy-making decisions, specifically in response to crises like COVID-19.

This paper investigates strategies to mitigate the spread of adversarial content on social networks using game theory and optimization techniques.

Academic Service

◦ Conference Reviewer:
    UAI 2023
    SecureComm 2023
    AAMAS 2023 (sub)
    NeurIPS 2024
    AAAI 2025

    ICLR 2025 (Upcoming)
    AISTATS 2025 (Upcoming)

◦ Workshop Reviewer:
    NLP4Science at EMNLP 2024
    ICML 2024 FM-Wild

◦ Journal Reviewer:
    DMLR

◦ Volunteer:
    UAI 2023