Talks

Last year, I presented AutoFR, a reinforcement learning framework that automatically generates filter rules for adblocking optimized per-site. This year, I extend this work and provide a comparative analysis of various new methodologies to generate filter rules that can generalize across multiple sites.

The talk presents our work on AutoFR for filter rule rule generation. We will walk through the formulation, implementation and evaluation of AutoFR.

The talk discusses how we can model the problem of filter rule generation as a reinforcement learning problem. We implement our RL framework and reveal some preliminary results.

The talk discusses the differences between different circumvention techniques (cloaking vs. obfuscation techniques). It will go over the state of anti-circumvention and how our work can help reduce human labeling effort to identify sites with successful circumvention.

After an overview of different circumvention techniques, this talk will describe an ML approach to automatically detect circumvention in the wild as well as identify popular third-party circumvention providers.