William Melicher

2 previous affiliations
5Machine learning papers tracked by Lacuna
Is this you?

Research Overview

William Melicher focuses on web security, usable privacy, and password authentication. His research trajectory spans the empirical evaluation of client-side vulnerabilities like DOM XSS, the usability of text passwords on mobile devices, and user perceptions of online tracking. Recently, his work investigates the application of machine learning to model password guessability and detect web application vulnerabilities.

Research Trajectory

William Melicher's research addresses critical challenges in web security, usable privacy, and authentication systems. As a lead author across several distinct but related domains, his work combines empirical measurement, human-computer interaction methodologies, and machine learning to evaluate and improve security mechanisms.

In the domain of web application security, Melicher has focused extensively on client-side vulnerabilities, particularly Document Object Model Cross-Site Scripting (DOM XSS). In Riding out DOMsday: Toward Detecting and Preventing DOM Cross-Site Scripting, he led an empirical assessment of DOM XSS prevalence on the web. By embedding taint tracking directly into a browser's JavaScript engine, this work collected a comprehensive dataset of potential vulnerabilities, improving upon prior crawling methodologies. Recognizing that dynamic taint tracking is often too computationally expensive for practical, widespread deployment, Melicher subsequently explored alternative detection mechanisms. This led to Towards a Lightweight, Hybrid Approach for Detecting DOM XSS Vulnerabilities with Machine Learning, where he and his collaborators trained deep neural networks on over 18 billion JavaScript functions to predict vulnerabilities, investigating whether machine learning classifiers could effectively augment or replace traditional taint tracking.

Parallel to his work on web vulnerabilities, Melicher has investigated usable privacy, specifically how users perceive and interact with online tracking. In (Do Not) Track Me Sometimes: Users’ Contextual Preferences for Web Tracking, he conducted a detailed study of the perceived risks and benefits of web tracking. Rather than studying these preferences in the abstract, the research grounded interviews in users' actual browsing histories, revealing that while users desire more control and recognize some benefits to controlled tracking, they are largely unwilling to expend the effort required to manage it manually.

Melicher has also made significant contributions to the study of password authentication, examining both the human factors of password creation and the adversarial modeling of password strength. Addressing the shift from desktop to mobile computing, he led the research in Usability and Security of Text Passwords on Mobile Devices. This study compared password strength and usability across devices, finding that creating passwords on mobile platforms is significantly more time-consuming, error-prone, and frustrating for users, which has direct implications for password policy design.

Current Directions

The most recent papers in the evidence indicate an ongoing focus on applying machine learning architectures to solve traditional security bottlenecks, particularly in authentication. Recent work investigates the use of artificial neural networks to model text passwords' resistance to guessing attacks. In Fast, Lean, and Accurate: Modeling Password Guessability Using Neural Networks, Melicher explores how different network architectures and training methods impact guessing effectiveness. This direction aims to replace older, computationally heavy models—such as probabilistic context-free grammars—with neural networks that are fast and lean enough for real-time, client-side password strength checking, demonstrating a broader interest in making advanced security evaluations practical for end-user devices.

Profile updated

See something wrong? Report a mistake