Dimensionality reduction for images of IoT using machine learning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Internet of Things

Background:

• Internet of Things (IoT) devices generate vast data, increasing network traffic and latency.
• Edge computing processes data closer to the source, mitigating cloud-related issues.
• Machine learning on the edge is crucial for efficient IoT data handling.

Purpose of the Study:

• To explore the integration of cloud and edge computing for IoT environments.
• To investigate machine learning methods for edge-based image dimensionality reduction.
• To evaluate the impact of data reduction on cloud-based machine learning tasks.

Main Methods:

• Utilized autoencoder deep learning and Principal Component Analysis (PCA) for image dimensionality reduction on the edge.
• Encoded data was transmitted to cloud servers for subsequent machine learning tasks.
• Evaluated the approach on an object detection task using 4000 images from COCO, human detection, and HDA datasets.

Main Results:

• A 77% reduction in data volume was achieved through edge processing.
• The significant data reduction did not substantially impact the accuracy of the object detection task.
• This demonstrates the feasibility of edge-based dimensionality reduction for IoT.

Conclusions:

• Merging cloud and edge computing with ML-driven dimensionality reduction is effective for IoT.
• Edge computing, using techniques like autoencoders and PCA, optimizes data processing for IoT applications.
• This strategy balances data reduction with maintained accuracy for cloud-based ML tasks.