The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Machine Learning for Security and Security for Machine Learning with Nicole Nichols - TWiML Talk #252
Apr 16, 2019
Nicole Nichols is a senior research scientist at Pacific Northwest National Lab, specializing in the intersection of machine learning and security. In this discussion, she shares insights from her presentation on using machine learning for cybersecurity, including detecting insider threats through advanced language models. The conversation also dives into software fuzz testing enhanced by deep learning, issues with algorithm efficiency, and the vulnerabilities in object recognition systems, highlighting the critical balance between detection accuracy and human analysis.
AI Snips
Chapters
Transcript
Episode notes
From Marine Mammals to Machine Learning
- Nicole Nichols's PhD focused on marine mammal detection and classification using human speech technologies.
- This involved determining species presence and even individual recognition.
Cybersecurity Use Cases
- Insider threat detection uses network data to find early indicators of malicious activity, like data exfiltration.
- Intrusion detection uses similar data, focusing on NetFlow and PCAP data to monitor network activity.
Fuzz Testing with Deep Learning
- Software fuzz testing explores code paths to find vulnerabilities by inputting random byte strings.
- Traditional methods use genetic algorithms, but deep learning models like GANs and LSTMs can accelerate this process.