Quantum-Inspired Tensor Network Algorithms for Efficient Large-Scale Data Compression

Abstract

The explosive expansion of data through digital architecture requires entirely new approaches in data compression that will be able to exceed limitations posed by conventional methods. This work introduces a new paradigm, namely Tensor Network Quantum-Inspired (TN-QI) Compression, that utilizes concepts from quantum computing, such as matrix product states (MPS) and multi-scale entanglement renormalization ansatz (MERA), within a classical computational context in order to achieve large-scale data compression. Using tensor networks as representations of high dimensional data and taking advantage of their intrinsic hierarchical correlations, TN-QI achieves a compression ratio of up to 31.8:1, outperforming conventional codecs such as HEVC and BPG. This methodology combines the techniques of adaptive bond dimension tuning, entanglement entropy-based truncation, and an encoding pipeline inspired by quantum circuits but that can be computed entirely on classical computers without any quantum coprocessor. Experiments performed on ImageNet-1K, 4K videos, and high-throughput genomics have shown improvements ranging from 40 to 60 percent in compression efficiency relative to classical algorithms, while also achieving a 35 percent lower encoding latency relative to comparable tensor decomposition methods. The approximation guarantees and computational complexity of the TN-QI algorithm were analyzed theoretically as well. These findings show that quantum-inspired tensor networks offer a promising, hardware-independent solution for future data compression systems.