[Colloquium] Andrew Chu MS Presentation - 11/09/2023

[Devin Davis]

This is an announcement of Andrew Chu's MS Presentation
===============================================
Candidate: Andrew Chu

Date: 11/09/2023

Time: 2:00 PM- 3:15 PM

Location: JCL 390

M.S. Paper Title: Transformer Models for Protocol Analysis: A Case Study in TLS

Abstract: Achieving generalizable, detailed methods for protocol analysis is difficult: Existing tools and approaches are not protocol-agnostic, and are typically specifically tailored for analyzing a specific protocol. Further, for sub-tasks of protocol analysis such as detecting and mitigating misconfiguration, tools detect errors only individually, and cannot provide insight into common patterns across protocol deployments that cause misconfiguration at large. In this paper, we perform a case study using transformer models for protocol analysis, specifically towards detecting and mitigating misconfiguration in the Transport Layer Security (TLS) protocol, examining if these models can (1) effectively detect TLS misconfiguration and (2) provide a different perspective towards more generalizable and proactive reasoning of errors in TLS implementation. To do so, we train a BERT-based model that learns semantically meaningful numerical representations (embeddings) for sites' TLSserver configurations. The model we develop contextualizes the fine-grained differences in various implementations of TLS, achieving 95+% accuracy for standard BERT pre-training tasks. We then fine-tune our model to classify (1)properly configured and misconfigured TLS implementations and (2) the reasons for misconfiguration; we also visualize the resulting prediction embeddings. We observe distinct clusters in these visualizations, and verify that these clusters indeed represent different reasons for proper configuration, or misconfiguration.

Advisors: Nick Feamster

Committee Members: Sanjay Krishnan, Junchen Jiang


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20231026/7cea0149/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: andrewcchu_ms.pdf
Type: application/pdf
Size: 657914 bytes
Desc: andrewcchu_ms.pdf
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20231026/7cea0149/attachment-0001.pdf>