Machine Learning Assisted Compression: Using Autoencoders for Image Compression on Commodity Hardware (A Case Study)

The incredible expanse of the Internet’s utility for commerce, entertainment, and social networking over the past 25 years is quite impressive. There is little doubt that an internet connection is a requirement to function in modern society. As our dependence on the internet grows, new innovations must be created to keep up with the increasing demand for our network infrastructure. Such innovative improvements on existing software algorithms find their implementation is far easier to deploy and scale than that of their hardware counterparts.

This paper takes a case-study approach for creating one such algorithmic improvement on an existing solution for file compression using recently researched methodologies in Machine Learning (ML). By employing the use of a neural network autoencoder, lossy image compression could be improved by compressing images through the autoencoder, transmitting the compressed binary file, and reconstructing the encoded binary on the target device. This may prove to be a more efficient way to distribute unstructured data such as images, audio, or video, etc. The motivation for creating such a system would be to improve the user experience on devices that have a high probability of network packet loss. These types of devices are often commodity cell phones on a 3G or 4G network.