This article proposes a method to generate accurate and concise analytic equations for device compact modeling using Physics-Assisted Kolmogorov–Arnold Networks (PKAN). The equations are directly extracted from the trained neural network architecture. PKAN uses variable activation functions informed by prior physical knowledge to model device behaviors. Similarity constraints map these trained activation functions to mathematical symbols. Sparsification techniques simplify the network structure, producing concise and explicit equations. This article also presents four approaches for physics-assisted device modeling using PKAN: (1) generating entire continuous equations without human intervention, (2) applying correlation factors to existing models without requiring knowledge of internal physical mechanisms, (3) revising specific parts of existing models, and (4) automatically extending existing models. Experimental results show that PKAN demonstrates significant accuracy improvements, achieving error reductions of 91.8%, 91.5%, 66.2%, and 83.7% for corresponding experiments, respectively. These findings demonstrate PKAN’s potential for various device modeling applications. By combining the precision of neural networks with the clarity of symbolic representation, PKAN offers a powerful tool for device modeling applications.

Accurately predicting phenotypes from genotypes holds great promise to improve health management in humans and animals, and breeding efficiency in animals and plants. Although many prediction methods have been developed, the optimal method differs across datasets due to multiple factors, including species, environments, populations and traits of interest. Studies have demonstrated that the number of genes underlying a trait and its heritability are the two key factors that determine which method fits the trait the best. In many cases, however, these two factors are unknown for the traits of interest. We developed a cloud computing platform for Mining the Maximum Accuracy of Predicting phenotypes from genotypes (MMAP) using unsupervised learning on publicly available real data and simulated data. MMAP provides a user interface to upload input data, manage projects and analyses and download the output results. The platform is free for the public to conduct computations for predicting phenotypes and genetic merit using the best prediction method optimized from many available ones, including Ridge Regression, gBLUP, compressed BLUP, Bayesian LASSO, Bayes A, B, Cpi and many more. Users can also use the platform to conduct data analyses with any methods of their choice. It is expected that extensive usage of MMAP would enrich the training data, which in turn results in continual improvement of the identification of the best method for use with particular traits. AVAILABILITY AND IMPLEMENTATION: The MMAP user manual, tutorials and example datasets are available at http://zzlab.net/MMAP. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.