A Constraints-Embeded OBS Process and Conjugate Gradient Optimizing Algorithm

The optimal brain surgeon (OBS) process is a high computing complexity post-training network pruning algorithm. A OBS conjugate gradient structure optimizing algorithm (c-OBS-CG) constraints embedded was designed which regarded the structure evaluating relationship as constraints and introduced it into neural network trainging target function, and aided by the mix process of Newton gradient and conjugate gradient. The simulation experiment verify its validity. It implements the parallel pruning of network training and OBS process, keeps both the accuracy of OBS procedure and highly-efficient regularization, raises the generalization performance of neural network.