115.05.21 演講公告
資工系專題演講(Lecture)
日期(date)/時間(time)：115年5月21日(四/Thu)13:10~15:00
地點(location)：電綜大樓B107教室
演講者(speaker)：黃俊穎 教授
服務單位(job)：國立陽明交通大學資訊工程學系
講題(topic)：State-Aware Fuzzing for Complex Network Protocols
摘要(summary)：
Modern network protocols rely on complex, multi-step interactions where behavior depends heavily on conversational context. Traditional fuzzing methods often treat inputs as independent messages or shallow sequences, limiting their ability to explore deep protocol logic and uncover interaction-dependent vulnerabilities such as state confusion, logic flaws, and sequence-triggered crashes.
In this talk, we present a fuzzer-in-the-middle architecture that improves protocol fuzzing without requiring explicit state modeling. Instead of maintaining internal state, the system passively intercepts and observes real client–server communications, extracting behavioral signals directly from live protocol exchanges.
We introduce flexible state selection strategies that guide fuzzing toward meaningful points in a conversation. These selections can be performed manually by an analyst or automatically using heuristics based on message patterns, response behavior, and coverage feedback. It allows targeted mutation and replay at semantically important stages. We demonstrate our approach using real-world examples from the 5G core network and the Bluetooth Low Energy (BLE) protocol, showing improved exploration of protocol logic and increased vulnerability discovery compared to baseline fuzzers.
與工程認證核心能力關聯性：
認識時事議題，瞭解資訊工程技術對環境、社會及全球的影響
日期(date)/時間(time)：115年5月21日(四/Thu)13:10~15:00
地點(location)：電綜大樓B107教室
演講者(speaker)：黃俊穎 教授
服務單位(job)：國立陽明交通大學資訊工程學系
講題(topic)：State-Aware Fuzzing for Complex Network Protocols
摘要(summary)：
Modern network protocols rely on complex, multi-step interactions where behavior depends heavily on conversational context. Traditional fuzzing methods often treat inputs as independent messages or shallow sequences, limiting their ability to explore deep protocol logic and uncover interaction-dependent vulnerabilities such as state confusion, logic flaws, and sequence-triggered crashes.
In this talk, we present a fuzzer-in-the-middle architecture that improves protocol fuzzing without requiring explicit state modeling. Instead of maintaining internal state, the system passively intercepts and observes real client–server communications, extracting behavioral signals directly from live protocol exchanges.
We introduce flexible state selection strategies that guide fuzzing toward meaningful points in a conversation. These selections can be performed manually by an analyst or automatically using heuristics based on message patterns, response behavior, and coverage feedback. It allows targeted mutation and replay at semantically important stages. We demonstrate our approach using real-world examples from the 5G core network and the Bluetooth Low Energy (BLE) protocol, showing improved exploration of protocol logic and increased vulnerability discovery compared to baseline fuzzers.
與工程認證核心能力關聯性：
認識時事議題，瞭解資訊工程技術對環境、社會及全球的影響