High Accuracy Preserving Regression-Based Physics Inversion Workflow Deployment on 8-Bit Integer Computing Hardware

Deep neural networks (DNNs) enable real-time regression-driven inversion in edge AI systems across various domains, including geophysical exploration. However, quantizing such models to low-bit precision for fast, power-efficient embedded deployment is challenging, especially in continuousvalued regression tasks that demand high accuracy. In this work, we deploy a multitask learning (MTL) DNN model on Neural Processing Unit (NPU) for predicting geophysical properties (e.g formation density, photoelectric factor, and mud properties). By combining per-layer quantization-aware training (QAT) and noise-augmented synthetic data, we achieve near-floating-point accuracy while drastically reducing inference latency. Experimental results show over <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$17 \times$</tex> speedup on the NPU compared to an inversion algorithm, a robust physics-based method currently in production at our company that, despite its high accuracy, cannot be deployed in real time.