Researchers Develop Novel AI Solution for Multi-Domain Applications

A team of researchers from Madhav Institute of Technology and Science has made significant strides in developing a general artificial intelligence solution that can solve different problems across various domains without the need for re-training. The breakthrough is the result of a unique approach that combines machine learning and genetic algorithms to generate high-performance convolutional neural network architectures. This innovative method, known as Genetic Algorithm-Based Search Space Exploration to Generate Best Convolutional Neural Network (GASE-BCNN), has demonstrated impressive results on benchmark datasets, including the German traffic signal classification dataset and CIFAR-10.

Key Takeaways:

• The researchers have developed a novel AI solution that can provide at least or above-average human-level performance in multiple domains without re-training for each domain separately.
• The solution combines machine learning and genetic algorithms to generate high-performance convolutional neural network architectures.
• The GASE-BCNN method enables the exploration of the hyperparameter search space more thoroughly, resulting in faster convergence and reduced training time and computations.
• The proposed method has been tested on the German traffic signal classification dataset and CIFAR-10, with results showing high fitness scores of up to 92% and 98%, respectively.
• The composite fitness metric, which integrates Validation Accuracy, F1-Score, and Regularization into the cross-entropy loss metric, has been found to improve model generalization, especially on imbalanced data.
• The researchers have successfully generated high-performance convolutional neural network architectures for search space exploration using the Genetic algorithm method.
• The proposed method has the potential to be applied to various multi-domain applications, including aerospace research and space exploration.

Statistics:

• The German traffic signal classification dataset and CIFAR-10 were used as benchmark datasets to test the proposed method.
• The GASE-BCNN method achieved high fitness scores of up to 92% and 98% on the German traffic signal classification and CIFAR-10 datasets, respectively.
• The composite fitness metric achieved meaningful improvements in model generalization, especially on imbalanced data.
• The proposed method converged faster compared to other neural architectural search methods, resulting in reduced training time and computations.
• The researchers demonstrated that the proposed method can be used to generate high-performance convolutional neural network architectures for search space exploration.

Sources:

• Genetic Algorithm-Based Search Space Exploration to Generate Best Convolutional Neural Network (GASE-BCNN). IEEE Access, 2025,13():155482-155499.
• Dhananjay Bisen, Madhav Institute of Technology and Science, Gwalior, Madhya Pradesh, India.
• Praneet Saurabh, Mayank Thakur, Gyanendra Chaubey, Upendra Singh, Aditya Dubey.