Abstract Optimizing neural network architectures through effective pruning techniques has become essential to balancing model complexity and accuracy. This study introduces a novel correlation-based approach to systematically reduce network size by identifying and removing redundant neurons based on their activation correlations. By selectively pruning neurons while compensating for their contributions, the method maintains model fidelity across diverse datasets. Results demonstrate substantial architecture reductions with minimal performance impact: For the MNIST dataset, the number of neurons in hidden layers was reduced from 128-128 to 118-93, while maintaining a high accuracy of 97.59%. Comparative analysis indicates that this pruning approach achieves competitive or superior results compared to state-of-the-art methods while reducing computational complexity and memory requirements by up to 25%. The findings highlight the potential of correlation-driven pruning strategies to optimize neural networks, making them more efficient and adaptable to resource-constrained environments.

A boundary gradient search technique is introduced. The technique generates a sequence of points on the boundary of the feasible region converging to a fixed point. The gradient of the boundary at this fixed point has a prespecified directional ratio. The boundary search technique is exploited in a modification of the ellipsoidal technique for design centering. This modification allows the use of double-sided ellipsoidal sections instead of single-sided ellipsoidal sections. It improves the speed of convergence of the ellipsoidal technique. Also, a computationally inexpensive technique of determining the gradient of the gain constraints is introduced. Practical examples are given to demonstrate the new technique.

A technique for design centering and feasible region approximation that is based on generating a sequence of ellipsoids of decreasing volume and preserves the property of containing a bounded convex feasible region is introduced. The technique converges to an ellipsoid the center of which is the proposed design center. The ellipsoid matrix can be used to give what is called a preferable covariance matrix, assuming a multinormal distribution of parameters. This covariance matrix is preferred since it significantly increases the production yield for the feasible region under consideration. In addition, an ellipsoidal region approximation can be obtained by scaling the final ellipsoid, which allows an inexpensive yield estimate using the Monte Carlo method. Numerical and practical examples are considered.>