Parallelizing Backpropagation Neural Network Using MapReduce and Cascading Model

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Artificial Neural Networks (ANNs), particularly backpropagation neural networks (BPNNs), excel at pattern recognition, classification, and prediction.
• Standard BPNNs face efficiency challenges with large datasets due to extensive computations.
• Traditional parallelization methods often compromise the precision of BPNNs.

Purpose of the Study:

• To develop a parallelized BPNN that enhances efficiency for large-scale machine learning.
• To address the trade-off between efficiency and precision in parallelized BPNNs.
• To introduce a novel cascading model for improved classification accuracy.

Main Methods:

• Implementation of a parallelized BPNN leveraging the MapReduce computing model.
• Integration of fault tolerance, data replication, and load balancing features within the MapReduce framework.
• Development of a cascading model-based classification approach to refine results.

Main Results:

• The parallelized BPNN demonstrates significantly improved efficiency in handling large datasets.
• The cascading classification approach effectively maintains or enhances classification precision.
• Experimental results validate the model's ability to balance efficiency and accuracy.

Conclusions:

• The MapReduce-based parallelized BPNN offers a scalable solution for large-scale machine learning tasks.
• The proposed cascading model successfully mitigates accuracy loss associated with parallelization.
• This approach provides a robust method for achieving high efficiency and precision in BPNN applications.