A Linear Algebra–Based Mathematical Model of Digital System Performance Using Neural Networks as Predictive Approximators

Title: A Linear Algebra-Based Mathematical Model of Digital System Performance Using Neural Networks as Predictive Approximators Authors: Goutam Parashuram Gotur, Chaitra Shree Affiliation: Department of Computer Science and Engineering, The Oxford College of Engineering, Bengaluru, India Abstract: This paper, authored by Goutam Parashuram Gotur and Chaitra Shree, introduces a hybrid mathematical framework for modeling digital system performance by combining a linear algebraic baseline with a neural network residual. The linear component captures interpretable, resource-level contributions, while the neural network models nonlinear effects such as cache behavior, queueing, and contention. The formulation includes error bounds, complexity analysis, and training strategies. Visual diagrams illustrate the hybrid architecture, and case studies demonstrate its practical relevance in CPU scheduling and memory bandwidth prediction. The model balances interpretability, accuracy, and computational efficiency, making it suitable for real-world deployment. Keywords: performance modeling, linear algebra, neural networks, hybrid models, predictive approximators, TikZ, digital systems, complexity analysis