Benchmarks & use cases
Logic-Based Networks versus the AI industry's best.
From energy consumption to speed, Literal Labs' unique, logic-based approach to AI offers key advantages over other AI algorithms. To give you a deeper understanding of performance and capabilities of LBNs, we occasionally publish new benchmarks and details on new use-cases.
A 14 kB model predicting sewage overflows in 1.74 ms, running on a battery-powered IoT sensor for over a decade without intervention. The neural network alternative couldn't even deploy to the target hardware.
Learn more: Sewage spills, predicted before they happen
Tested against the MLCommons MLPerf Tiny benchmark, Literal Labs' logic-based AI ran 54× faster and consumed 52× less energy than the best published neural network result — on identical hardware.
Learn more: Anomaly detection with Tsetlin Machines and MLPerf Tiny
Trained on Hitachi's real-world industrial MIMII dataset, Literal Labs' LBN outscored a standard neural network on both accuracy and F1 — proving logic-based AI on factory-floor conditions.
Learn more: Pump sounds and anomaly detection with Tsetlin Machines
Benchmarked across seven datasets on an ESP32 microcontroller, Literal Labs' logic-based AI ran 250× faster than XGBoost whilst using 130 kB less memory — with broader applicability and full explainability.
Learn more: Tsetlin Machine versus XGBoost
