deforce: Derivative-free algorithms for optimizing Cascade Forward Neural Networks
Nguyen Van Thieu, Hoang Nguyen, Harish Garg, Gia Sirbiladze · Software Impacts · 2024
This paper aims to introduce the ‘deforce’ framework, an open-source Python library constituted on top of Numpy, Scikit-Learn, PyTorch, and Mealpy. This framework provides hybrid models that combine derivative-free techniques with Cascade Forward Neural Networks (CFNNs). By inheriting from scikit-learn’s estimator, deforce’s models ensure easy integration into existing machine learning pipelines. It also has many advantages, including a simple installation process, a user-friendly interface, and adaptability to various user requirements. For researchers and practitioners looking to improve CFNN performance with minimal implementation effort, deforce offers a useful and approachable option. • A unified Python framework for Cascade Forward Neural Network-based (CFNN) Models. • Derivative-Free Optimization (DFO) is employed to create hybrid models. • DFO is utilized to optimize the weights of CFNN models. • DFO is employed to optimize the hyperparameters of CFNN models.