TextEconomizer: Enhancing Lossy Text Compression with Denoising Autoencoder and Entropy Coding
Keywords: Text Compression, Denoising AutoEnccoder, Lossy Text, Entropy Coding, Latent Space
TL;DR: TextEconomizer: Enhancing Lossy Text Compression with Denoising Autoencoder and Entropy Coding
Abstract: Lossy text compression reduces data size while preserving core meaning, making it ideal for summarization, automated analysis, and digital archives where exact fidelity is less critical. While extensively used in image compression, text compression techniques, such as integrating entropy coding with autoencoder latent representations in Seq2Seq text generation, have been underexplored. A key challenge is incorporating lossless entropy coding into denoising autoencoders to improve storage efficiency while maintaining high-quality outputs, even with noisy text. Prior studies have mainly focused on near-lossless token generation with little attention to space efficiency. In this paper, we present a denoising autoencoder with a rectified latent representation that compresses variable-sized inputs into a fixed-size latent space without prior knowledge of dataset dimensions. By leveraging entropy coding, our model achieves state-of-the-art compression ratios alongside competitive text quality, as measured by diverse metrics. Its parameter count is approximately 196 times smaller than comparable models. Additionally, it achieves a compression ratio of 67× while maintaining high BLEU and ROUGE scores. This significantly outperforms existing transformer-based models in memory efficiency, marking a breakthrough in balancing lossless compression with optimal space optimization.
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 11419
Loading