Data Driven Security
summary
Claims that machine learning plus human labeling cuts analysis effort by 8x fewer resources compared with manual decompilation, using sandbox traces → metadata → a second classifier to turn probabilistic scores into actionable blocks. Uses VGG-16 vectorization for raw streams, NLP-style vectors for API sequences, and a 300,000-dimensional DNA embedding to cluster ~60 malware families. Runs federated learning with supervised labels, flagged a ransomware sample days before others, and pushes continuous retraining into production.
AI-written summary · about
chapters (11)
- 0:00Start
- 7:39Attacking Is Easier Than Defending
- 11:07Changing Representation Turning an image into a number - VGG16 Convolutional Network
- 13:20Vectorising Elements - Example: Human Language
- 14:53Vectorising Elements - Natural Language Processing
- 15:19Vectorising Elements - Why is NLP useful?
- 16:27Vectorising Elements - Cyber Security
- 18:24Changing Representation Turning an executable file into vectors - VGG16 Convolutional Net
- 18:51Machine Learning In Cyber Security
- 26:32Machine Learning - General Purpose Comparing NLP-Trained Models
- 27:19Machine Learning Empowers Threat Prevention
related talks
32mAI/Machine Learning: What is Actually Working in Cybersecurity?
27mA Machine Learning approach for detecting a Malware
28mSecurity of Machine Learning
48mCatch me if you can: Building a Web Malware Analyzer using Machine Learning
12mMachine Learning in the DevSecOps Pipeline
31mBuild threat detection models using ML
27m"Cybertelligence": How AI is changing the face of cybersecurity
44mAdvancing Cybersecurity with AI and Machine Learning: Opportunities and Challenges