Common Optimization of Adaptive Preprocessing Units and a Neural Network during the Learning Period. Application in...

Area of Science:

• Artificial Intelligence
• Biomedical Engineering
• Signal Processing

Background:

• Neonatal electroencephalogram (EEG) monitoring requires efficient signal processing for accurate analysis.
• Traditional neural network training methods may not be optimal for complex biological signals like neonatal EEG.
• Adaptive preprocessing is crucial for handling variations in biological data.

Purpose of the Study:

• To propose a novel method for simultaneous training of neural networks and adaptive preprocessing units.
• To enhance the convergence and efficiency of neural network learning for pattern vector separation.
• To enable efficient and reliable EEG monitoring in neonates.

Main Methods:

• Simultaneous training of a multilayer perceptron neural network and an adaptive preprocessing unit.
• Allowing the neural network to influence preprocessing for dynamic feature space manipulation.
• Developing a strategy to optimize pattern vector distribution for improved class separation.

Main Results:

• The cooperative training strategy leads to improved convergence of the learning process.
• The method effectively varies pattern vector locations in the feature space for better separation.
• Demonstrated potential for efficient neonatal EEG monitoring.

Conclusions:

• The presented simultaneous training approach is a viable alternative to conventional neural network learning strategies.
• The cooperative system ensures reliable convergence for complex signal processing tasks.
• This method holds promise for advancing neonatal neurological monitoring.