Academic Research

Investigating LLM vulnerabilities and building defensive systems against adversarial attacks

Research Focus

My research centers on enhancing the security and reliability of large language models (LLMs), especially in adversarial settings. I study how subtle modifications in scam messages can exploit LLM vulnerabilities and lead to misclassifications. To address this, I develop adversarial benchmarks, structured perturbation techniques, and hybrid detection systems that combine fine-tuned LLMs with traditional machine learning models. These contributions aim to strengthen the robustness of LLM-based systems against real-world manipulative attacks.

During my 2025 internship with TSMC’s AI R&D group, I applied these ideas to high-volume semiconductor manufacturing by deploying Qwen3 models on NVIDIA H100 GPUs and automating anomaly detection workflows. The real-world feedback loop from that work continues to inform my research directions.

Key research areas include:

  • Adversarial attacks against large language models
  • Detection and prevention of AI-enabled scams
  • Ensemble methods for improving model robustness
  • Fine-tuning techniques for defensive capabilities

Academic Publications

Scam Shield: Multi-Model Voting and Fine-Tuned LLMs Against Adversarial Attacks

Chen-Wei Chang, et al.

2025 IEEE International Conference on Big Data (IEEE BigData 2025), Macau, China

Published

Abstract

Scam detection remains a critical challenge in cybersecurity, especially with the increasing sophistication of adversarial scam messages designed to evade detection. This work proposes a Hierarchical Scam Detection System (HSDS) that integrates multi-model voting with a fine-tuned LLaMA 3.1 8B Instruct model to improve detection accuracy and robustness against adversarial attacks. Our approach leverages a four-model ensemble for initial scam classification, where each model independently evaluates scam messages, and a majority voting mechanism determines preliminary predictions. The final classification is refined using a fine-tuned LLaMA 3.1 8B Instruct model, optimized through adversarial training to mitigate misclassification risks. Experimental results demonstrate that our hierarchical framework significantly enhances scam detection performance, surpassing both traditional machine learning models and larger proprietary LLMs, such as GPT-3.5 Turbo, while maintaining computational efficiency. The findings highlight the effectiveness of a hybrid voting mechanism and adversarial fine-tuning in fortifying LLMs against evolving scam tactics, enhancing the resilience of automated scam detection systems.

Contributions

  • Proposed a hybrid detection system combining four ML models with a fine-tuned LLaMA 8B
  • Introduced efficient LoRA-based fine-tuning for adversarial robustness
  • Created a 20K-sample scam dataset with adversarial examples
  • Benchmarked detection accuracy across various scam types

Impact

  • Boosts scam detection accuracy while reducing computational cost
  • Enhances LLM resilience to evolving adversarial attacks
  • Enables practical deployment of AI security tools
  • Informs future research on robust ensemble-based detection methods

Exposing LLM Vulnerabilities: Adversarial Scam Detection and Performance

Chen-Wei Chang, et al.

2024 IEEE International Conference on Big Data, Washington, DC, USA

Published

Abstract

Can we trust Large Language Models (LLMs) to accurately predict scam? This paper investigates the vulnerabil- ities of LLMs when facing adversarial scam messages for the task of scam detection. We addressed this issue by creating a comprehensive dataset with fine-grained labels of scam messages, including both original and adversarial scam messages. The dataset extended traditional binary classes for the scam detection task into more nuanced scam types. Our analysis showed how adversarial examples took advantage of vulnerabilities of a LLM, leading to high misclassification rate. We evaluated the performance of LLMs on these adversarial scam messages and proposed strategies to improve their robustness.

Contributions

  • Created a fine-grained labeled dataset with original and adversarial scam messages
  • Designed a structured method to generate adversarial examples using prompt engineering
  • Benchmarked LLMs like GPT-3.5, Claude3, and LLaMA on scam detection across multiple categories
  • Identified performance degradation patterns under adversarial settings

Impact

  • Revealed key vulnerabilities of LLMs to adversarial scam messages
  • Highlighted the need for adversarial training to improve model robustness
  • Provided practical guidelines for evaluating LLMs in security-critical tasks
  • Contributed new insights to the field of adversarial robustness in NLP

RailEstate: An Interactive System for Metro Linked Property Trends

Chen-Wei Chang, Yu-Chieh Cheng, Yun-En Tsai, Fanglan Chen, Chang-Tien Lu

33rd ACM International Conference on Advances in Geographic Information Systems (SIGSPATIAL '25), Minneapolis, MN, USA

Published

Abstract

RailEstate is a web-based analytics platform that fuses 25 years of Washington metropolitan housing data with WMATA transit infrastructure to surface metro-linked price trends. The system supports interactive spatial queries, time-series visualizations, forecasting, and a natural language to SQL chatbot that transforms plain-English questions into optimized PostGIS queries. By unifying spatial databases, forecasting pipelines, and LLM-powered interfaces, RailEstate delivers actionable, transit-aware housing insights for planners, investors, and residents without requiring technical expertise.

Contributions

  • Deployed a Supabase-hosted PostGIS stack with spatial indexing for low-latency metro proximity analytics
  • Integrated React/Leaflet visualizations and Recharts time-series views for interactive metro-centric price exploration
  • Built a LangChain and GPT-4o-mini text-to-SQL chatbot that executes validated natural language housing queries in real time

Impact

  • Enables non-technical audiences to interrogate transit-oriented development patterns instantly
  • Supports data-driven planning decisions with forecasts that capture infrastructure and economic shifts
  • Provides a portable architecture for geospatial housing intelligence in additional metro areas

Future Research Directions

My ongoing and future research agenda aims to address several critical challenges in AI security and LLM robustness:

I welcome collaboration opportunities in these research areas. Please contact me to discuss potential research partnerships.

Contact Me

wilsonchang@vt.edu
+15715947580
Alexandria, VA, USA