Markowitz-Informed Neural Networks (MINNs): An Interpretable Deep Learning Approach to Portfolio Optimization

This paper introduces Markowitz-Informed Neural Networks (MINNs), a novel framework that unites the principles of mean-variance optimization (MVO) with the adaptability and power of modern neural networks. By learning portfolio weights and interpretable covariance structure simultaneously, without matrix inversion, MINNs offer a transparent and more stable alternative to traditional MVO. Embedding financial structure within the learning process, they support practical, robust, and regulation-aligned portfolio management across problem scales. Empirical results on a diversified portfolio show consistent outperformance over conventional methods.