SIRAJ - A Unified Framework for Aggregation of Malicious Entity Detectors

Explore a conference talk on SIRAJ, a unified framework for aggregating malicious entity detectors. Delve into the challenges of internet threats and learn about a novel approach to combat them. Discover how this framework addresses varying accuracy and expertise of scanners, label flips in predictions, and scanner correlations. Understand the implementation of self-supervised learning techniques, including pretext tasks for learning temporal scanner dependencies and representation consistency. Examine the high-level and detailed overall approach of SIRAJ, and analyze its performance through evaluation metrics. Compare SIRAJ's effectiveness against baselines for early detection and different training sizes. Gain valuable insights into this innovative solution for enhancing cybersecurity and malicious entity detection.

Intro
Internet Threats
Our Approach: A Unified Aggregating Framework
Varying Accuracy and Expertise of Scanners
Label Flips in Scanner Predictions
Scanner Correlations
Self Supervised Learning
High-Level Overall Approach
Pretext Task 2: Learn Temporal Scanner Dependencies
Pretext Task 3: Representation Consistency
Detailed Overall Approach
Evaluation
Siraj vs. Baselines for Early Detection
Siraj vs. Baselines for Different Training Size
Summary