01-31, 10:05–10:25 (Europe/Zurich), Beacon Stage
The talk explores an intelligence-based transaction analysis model within the Ethereum Virtual Machine (EVM), providing insights into transaction behaviors and security threats. Designed to enhance transaction processing capabilities, the model leverages machine learning, a unique approach, and specialized data sets to fortify security intelligence.
By identifying and highlighting known malicious behaviors and exploit patterns, the approach aims to effectively mitigate security risks. Operating in two distinct modes, it offers real-time analysis for immediate threat detection and on-demand processing for in-depth investigation. The core strength lies in systematically storing generated intelligence in a dedicated database, enabling efficient retrieval and analysis.
We will present the applications of each mode, from historical data enrichment to smart simulations, and their implications for enhancing Ethereum's security ecosystem.
The talk focuses on machine learning-powered security model which transforms Ethereum transaction analysis by providing intelligent, context-aware threat detection. Leveraging advanced outlier detection across multiple transaction dimensions, we can identify potential security risks with high precision. The model operates through dual analysis modes: real-time threat detection and comprehensive investigative processing. By analyzing complex transaction patterns—including gas usage, address behaviors, and execution flows—we'll showcase how we can detect anomalies that traditional security tools miss. Our approach goes beyond binary risk flagging, offering nuanced, plain-language explanations of potential threats.
Unlike conventional security tools that provide opaque warnings, this approach delivers transparent, actionable intelligence. It transforms complex blockchain data into clear insights, enabling safer, more confident Web3 interactions.
The model's architecture allows for flexible application across various use cases, from immediate threat detection to retrospective security analysis. By systematically capturing and analyzing transaction intelligence, we're not just identifying risks, but instead building a comprehensive knowledge base that anticipates and mitigates emerging security challenges in the Ethereum ecosystem.
Early tests demonstrate real-world effectiveness, with successful identification of anomalous patterns that traditional methods overlook. This positions the model at the forefront of Web3 security, bridging the gap between sophisticated threat detection and user-friendly accessibility.
Aleksandar Kiridzic is a Principal Machine Learning Engineer at Tenderly. Prior to joining Tenderly and continuing his career in Web3, Aleksandar worked for Microsoft and Google in various roles, mostly focusing on ML functions and computer vision.
He holds an MSc in Electrical Engineering and Computer Science, and is currently pursuing a PhD degree in machine learning, with a focus on graph neural networks.
Aleksandar's passion for crunching numbers led him into the fields of statistics, data science and machine learning.