Know What's Happening in AI

Explainable AI Achieves 83.5% Accuracy with Quantized Active Ingredients and Boltzmann Machines

Researchers introduced a hybrid quantum‑classical framework that uses Quantized Boltzmann Machines (QBMs) to improve both performance and transparency in AI decision‑making. Tested on a binarized, PCA‑reduced MNIST subset, QBMs reached 83.5% classification accuracy, far surpassing the 54% of classical Boltzmann Machines (CBMs). The study employed gradient‑based saliency maps for QBMs and SHAP values for CBMs, revealing more concentrated feature‑attribution distributions (entropy 1.27 vs 1.39). These results suggest quantum‑enhanced models can deliver more trustworthy, explainable AI.