Adaptive Arithmetic Coding using Neural Network

In today’s data-driven world, industries across various sectors are grappling with the enormous influx of data being generated. The sheer volume of data poses significant challenges in terms of infrastructure costs, data transfer speeds, and storage requirements. As a result, many industries are reevaluating their approach to cloud-based storage solutions. One crucial solution to mitigate the issues associated with excessive storage usage, data integrity, transfer efficiency, and improved streaming speeds is lossless data compression. Numerous compressors currently available employ prediction-based compression techniques and utilize learned models to address these challenges effectively.To tackle this problem, our research focuses on exploring and developing strategies for compressing sequential data using neural network predictors. Neural networks are well-known as universal function approximators capable of learning complex mappings, and they have demonstrated impressive performance in prediction tasks. We leverage this capability by combining an arithmetic coder with recurrent neural network predictions, enabling us to achieve lossless compression for a wide range of datasets including synthetic, text, and genomic data [1]. Through a series of experiments using both simulated and actual text and genomic datasets, we evaluate the effectiveness of our approach. The results and observations from these experiments provide valuable insights for future research and advancements in the field of lossless data compression.

I. Introduction

The volume of data produced over the last few years has increased dramatically. New types of data, including as genomic [1], 3D VR, and point cloud data, are being produced at a rapid rate alongside picture and text data. Data production each year is increasing more quickly than ever before. Every person will produce 17 megabytes of information every second by the end of this year. There will be a greater need for storage as a result of the rapid development of digitalization, which also adds to the rapid accumulation of data. This means an increase in storage costs. The proliferation of multimedia files is evident from reports such as Cisco’s analysis of internet traffic, which predicts a staggering 2.75-fold increase in global traffic between 2016 and 2021 [2]. This exponential growth in data volume necessitates effective solutions to address the impact on download speeds, infrastructure costs, and end-user prices. Image compression plays a pivotal role in mitigating these challenges, as even small improvements in compression techniques can yield substantial benefits in terms of faster downloads, reduced infrastructure expenses, and more affordable services for end-users. Therefore, data compression stands as a critical solution to navigate the ever-expanding landscape of multimedia data.