Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Quantum Implicit Neural Compression

Takuya Fujihashi, Toshiaki Koike-Akino

TL;DR

The paper tackles the challenge of efficiently compressing multimedia signals with implicit neural representations by addressing the difficulty of preserving high-frequency details in small models. It proposes quINR, a hybrid quantum-classical INR that leverages a quantum neural network to increase representational expressivity and potentially reduce parameter counts. Across KITTI LiDAR RI and Kodak color images, quINR demonstrates improved rate-distortion performance, notably achieving up to 1.2 dB PSNR gains in certain regimes. The work highlights the potential of quantum models for data compression while acknowledging limitations in color-image scenarios and outlining future directions such as quantum neural architecture search and distillation to boost performance.

Abstract

Signal compression based on implicit neural representation (INR) is an emerging technique to represent multimedia signals with a small number of bits. While INR-based signal compression achieves high-quality reconstruction for relatively low-resolution signals, the accuracy of high-frequency details is significantly degraded with a small model. To improve the compression efficiency of INR, we introduce quantum INR (quINR), which leverages the exponentially rich expressivity of quantum neural networks for data compression. Evaluations using some benchmark datasets show that the proposed quINR-based compression could improve rate-distortion performance in image compression compared with traditional codecs and classic INR-based coding methods, up to 1.2dB gain.

Quantum Implicit Neural Compression

TL;DR

The paper tackles the challenge of efficiently compressing multimedia signals with implicit neural representations by addressing the difficulty of preserving high-frequency details in small models. It proposes quINR, a hybrid quantum-classical INR that leverages a quantum neural network to increase representational expressivity and potentially reduce parameter counts. Across KITTI LiDAR RI and Kodak color images, quINR demonstrates improved rate-distortion performance, notably achieving up to 1.2 dB PSNR gains in certain regimes. The work highlights the potential of quantum models for data compression while acknowledging limitations in color-image scenarios and outlining future directions such as quantum neural architecture search and distillation to boost performance.

Abstract

Signal compression based on implicit neural representation (INR) is an emerging technique to represent multimedia signals with a small number of bits. While INR-based signal compression achieves high-quality reconstruction for relatively low-resolution signals, the accuracy of high-frequency details is significantly degraded with a small model. To improve the compression efficiency of INR, we introduce quantum INR (quINR), which leverages the exponentially rich expressivity of quantum neural networks for data compression. Evaluations using some benchmark datasets show that the proposed quINR-based compression could improve rate-distortion performance in image compression compared with traditional codecs and classic INR-based coding methods, up to 1.2dB gain.
Paper Structure (14 sections, 2 equations, 4 figures)

This paper contains 14 sections, 2 equations, 4 figures.

Table of Contents

  1. Background
  2. Related Work
  3. Implicit Neural Compression
  4. Quantum Neural Network
  5. Quantum INR for Data Compression
  6. Encoding and Decoding Process
  7. Model Architecture
  8. Experiments
  9. Settings
  10. Datasets:
  11. Baseline:
  12. Implementation:
  13. Performance Comparison
  14. Conclusion

Figures (4)

  • Figure 1: Overview of the proposed scheme for data compression using hybrid quantum-classical implicit neural representation.
  • Figure 2: Exemplar architecture of QNN layer.
  • Figure 3: PSNR vs. bpp for RI.
  • Figure 4: PSNR vs. bpp for Kodak color image.