ABSTRACT
This survey paper provides a comprehensive overview of compression-aware machine learning (ML) and deep learning (DL) models. As the volume of information keeps on developing dramatically, the requirement for effective information processing and storage becomes increasingly critical. This paper explores the intersection of data compression techniques and ML/DL models, focusing on how these models can leverage compressed data representations to improve efficiency and performance. We review various compression techniques, their applications in ML and DL architectures, and the advantages of operating in the compressed domain. The survey covers recent advancements in feature extraction from compressed data, adaptation of ML models to the compressed domain, and specialized DL models designed for compressed inputs. Our findings highlight the potential of compression-aware approaches to address challenges in large-scale data processing and resource-constrained environments.
