Embed AI Anywhere
BareML™ creates the fastest and lightest-weight edge ML models possible using times-series data, ideal for IoT and Wearables.
Benchmarking Results for Anomaly Detection
Abstract
The ResCon team evaluates a BareML™-generated and trained Anomaly Detection model using the MLPerf Tiny benchmarking framework. Model accuracy exceeds the benchmark’s quality target while evaluating over 88,000x faster and using 41,000x less power than the MLPerf baseline model.
Market research reveals several entities that have also beaten the original benchmark by significant margins. We still find a 220x improvement in evaluation speed and a 310x improvement in energy consumption over the most efficient of these models.
BareML™ models achieve state-of-the-art accuracy while displaying ultra-fast training times on minimum data. The results presented in the white paper arise from an algorithm training time of only milliseconds on standard laptop hardware. This shows that BareML™ enables local accomplishment of the entire Machine Learning model workflow on non-specialized hardware—even directly on the target microcontroller.