Presentations
Talks & Podcasts
Conference presentations, invited talks, and podcast appearances.
Ad-Filtering Dev Summit 2024
Evaluating Large Language Models as a Defense Against Online Tracking
Exploring how LLMs can be leveraged to detect and block tracking JavaScript at the function level granularity, enabling fine-grained privacy protection while preserving website functionality.
USENIX Security 2024
PURL: Safe and Effective Sanitization of Link Decoration
Presenting a machine-learning approach that uses cross-layer graph representation of webpage execution to safely and effectively sanitize tracking information in decorated links.
IMDEA Networks
Beyond Third-Party Cookies: Safeguarding User Data from Storage and Exfiltration with CookieGraph and PURL
A comprehensive talk covering two complementary approaches to combat emerging tracking techniques: CookieGraph for first-party cookie tracking and PURL for link decoration tracking.
ACM CCS 2023
COOKIEGRAPH: Measuring and Countering First-Party Tracking Cookies
Presenting a machine learning-based approach that accurately detects and blocks first-party tracking cookies that are increasingly used as third-party cookies become blocked by browsers.
Ad-Filtering Dev Summit 2023
What you don't remove can track you: Measuring and detecting tracking decorations
Discussing how link decorations are abused by advertisers and trackers to exfiltrate user information, and how PURL can detect and sanitize these tracking decorations.
Ad-Filtering Dev Summit 2022
COOKIEGRAPH: Measuring and Countering First Party Tracking Cookies
Early presentation of CookieGraph research showing how first-party tracking cookies are used on 89.86% of top websites, with 96.61% being ghostwritten by third-party scripts.
DataSkeptic Podcast
First-Party Tracking Cookies
A podcast discussion explaining first-party tracking cookies, how they differ from third-party cookies, and the implications for user privacy as browsers block third-party cookies.
EACL 2021
Attribution of Text Generated by Language Models
Presenting machine learning methods to attribute synthetic text to its source language model, achieving 91-98% accuracy in identifying the parent pre-trained LM behind generated text.