HybridDNN-GA: A Genetic Algorithm to Optimize Layers of Hybrid Deep Neural Networks
Nileymegum Sivadilingam, Oomesh Gukhool · 2024
In recent years, the emergence of Hybrid Deep Learning Models (HDLMs) has been driven as a possible paradigm that merges the strengths of multiple neural network architectures to create a singular, more potent model. While the combination of models in HDLMs shows promising performance, it also introduces a complex NP-hard combinatorial challenge, given the number of combinations of possible architectures within the hidden layers. In this research we use a Genetic Algorithm (GA) to solve this NP-hard combinatorial problem inherent to the design of HDLMs. This study showcases the strength of a GA in crafting optimized HDLMs tailored to the specific problem and dataset (supervised learning) with a performance improvement of up to 10% in terms of prediction accuracy compared to other neural networks.