Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore a groundbreaking conference presentation that demonstrates how to enhance acoustic side-channel attacks on keyboards using advanced machine learning techniques. Learn about the integration of Visual Transformers and Large Language Models to overcome the limitations of current state-of-the-art models in noisy, real-world conditions. Discover how researchers from Texas A&M University and University of Toronto developed the first approach combining Visual Transformers for capturing long-term contextual information with LLMs for correcting keystroke prediction errors or "typos." Understand the methodology behind using transformer architectures to analyze audio signals from keyboard typing captured through device microphones, and examine how this dual-strategy approach addresses the challenge of inferring sensitive information from keystroke sounds in realistic noisy environments. Gain insights into the evaluation results showing that Visual Transformers achieve superior performance compared to previous CNN benchmarks, while LLMs like GPT-4o significantly boost error-correction capabilities, with comparable results achievable using lightweight, fine-tuned smaller models that are 67 times smaller than GPT-4o through Low-Rank Adaptation techniques.
