MapReduce Based Parallel Neural Networks in Enabling Large Scale Machine Learning

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Artificial neural networks (ANNs) are powerful tools for pattern recognition and classification.
• The computational demands of ANNs hinder their application with large datasets.
• Big data initiatives necessitate faster processing for machine learning models.

Purpose of the Study:

• To accelerate the computation of artificial neural networks (ANNs) for big data applications.
• To investigate the parallelization of ANNs using the MapReduce computing model.
• To evaluate the performance of parallelized ANNs under various data-intensive scenarios.

Main Methods:

• Parallelization of neural network computations using the MapReduce framework.
• Testing across three scenarios: varying classification data volume, training data size, and neuron count.
• Performance evaluation in a MapReduce computer cluster.

Main Results:

• Demonstrated significant speed-up in ANN computation for large datasets.
• Achieved comparable or improved classification accuracy with parallelized ANNs.
• Quantified efficiency gains in terms of computation time.

Conclusions:

• MapReduce-based parallelization effectively addresses the computational limitations of ANNs in big data contexts.
• The proposed approach enhances the feasibility of deploying ANNs for large-scale pattern recognition and classification tasks.
• Parallelized ANNs offer a viable solution for efficient big data processing.