Analytic Learning for Feed-Forward Deep Neural Network with Correlation Projection
Authors: Manu Pratap Singh, Pratibha Rashmi Department of Computer Science, Dr. Bhimarao Ambedkar University, Agra, UP, India Email:pratibha.rashmi@gmail.com
Journal: International Journal of Computer Techniques (IJCT)
Volume: 12 | Issue: 5 | Page: 35 | Publication Date: September – October 2025
This paper presents a novel feed-forward deep neural network architecture that replaces mini-batch stochastic gradient descent with analytic learning. Using a modified Moore-Penrose inverse and correlation projection, the model encodes labels into hidden layers and extracts features for improved classification. Experiments on spoken digit datasets show superior accuracy and reduced training time compared to traditional methods.
The proposed analytic learning framework with correlation projection outperforms conventional SGD-based models in classification accuracy and training efficiency. It reduces overfitting and enables modular learning with dimension-dependent label encoding. Future work may explore scalability across datasets and integration with hybrid learning strategies.
References
Includes 25+ references from NeurIPS, IEEE, Springer, MIT Press, and arXiv covering neural network learning algorithms, pseudo-inverse methods, and pattern recognition frameworks.
