CEEM: a Chemically Explainable Deep Learning Platform for Identifying Compounds with Low Effective Mass

The semiconductor industry occupies a crucial position in the fields of integrated circuits, energy, and communication systems. Effective mass (mE ), which is closely related to electron transition, thermal excitation, and carrier mobility, is a key performance indicator of semiconductor. However, the highly neglected mE is onerous to measure experimentally, which seriously hinders the evaluation of semiconductor properties and the understanding of the carrier migration mechanisms. Here, a chemically explainable effective mass predictive platform (CEEM) is constructed by deep learning, to identify n-type and p-type semiconductors with low mE . Based on the graph network, a versatile explainable network is innovatively designed that enables CEEM to efficiently predict the mE of any structure, with the area under the curve of 0.904 for n-type semiconductors and 0.896 for p-type semiconductors, and derive the most relevant chemical factors. Using CEEM, the currently largest mE database is built that contains 126 335 entries and screens out 466 semiconductors with low mE for transparent conductive materials, photovoltaic materials, and water-splitting materials. Moreover, a user-friendly and interactive CEEM web is provided that supports query, prediction, and explanation of mE . CEEM's high efficiency, accuracy, flexibility, and explainability open up new avenues for the discovery and design of high-performance semiconductors.

SALSA-Net: Explainable Deep Unrolling Networks for Compressed Sensing

Deep unrolling networks (DUNs) have emerged as a promising approach for solving compressed sensing (CS) problems due to their superior explainability, speed, and performance compared to classical deep network models. However, the CS performance in terms of efficiency and accuracy remains a principal challenge for approaching further improvements. In this paper, we propose a novel deep unrolling model, SALSA-Net, to solve the image CS problem. The network architecture of SALSA-Net is inspired by unrolling and truncating the split augmented Lagrangian shrinkage algorithm (SALSA) which is used to solve sparsity-induced CS reconstruction problems. SALSA-Net inherits the interpretability of the SALSA algorithm while incorporating the learning ability and fast reconstruction speed of deep neural networks. By converting the SALSA algorithm into a deep network structure, SALSA-Net consists of a gradient update module, a threshold denoising module, and an auxiliary update module. All parameters, including the shrinkage thresholds and gradient steps, are optimized through end-to-end learning and are subject to forward constraints to ensure faster convergence. Furthermore, we introduce learned sampling to replace traditional sampling methods so that the sampling matrix can better preserve the feature information of the original signal and improve sampling efficiency. Experimental results demonstrate that SALSA-Net achieves significant reconstruction performance compared to state-of-the-art methods while inheriting the advantages of explainable recovery and high speed from the DUNs paradigm.