High-Efficiency Lossy Source Coding Based on Multi-Layer Perceptron Neural Network

Area of Science:

• Computer Science
• Signal Processing
• Machine Learning

Background:

• Sensor networks generate vast data volumes, necessitating efficient lossy compression under bandwidth constraints.
• Conventional two-stage compression methods exhibit high computational complexity and struggle with balancing compression and generalization.

Purpose of the Study:

• To propose an end-to-end lossy compression method that overcomes limitations of traditional approaches.
• To enhance compression performance, generalization ability, and reconstruction quality in sensor networks.

Main Methods:

• Integration of an enhanced belief propagation algorithm with a multi-layer perceptron neural network.
• Development of a novel "encoding-structured encoding-decoding" joint optimization architecture.
• Implementation of a quantization module with random perturbation and straight-through estimator to handle non-differentiability.

Main Results:

• Significant improvements in compression performance and reconstruction quality were demonstrated.
• The proposed system exhibited superior generalization capabilities compared to conventional methods.
• The neural architecture proved simple and efficient, reducing overall system complexity.

Conclusions:

• The end-to-end lossy compression method offers a more effective and feasible solution for sensor network data.
• The novel joint optimization and quantization techniques contribute to enhanced compression efficiency and quality.
• The simplified architecture facilitates practical deployment in real-world applications.