![[2602.23012] Sequential Regression for Continuous Value Prediction using Residual Quantization](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.23012] Sequential Regression for Continuous Value Prediction using Residual Quantization
Summary
This article presents a novel approach to continuous value prediction using a residual quantization framework, enhancing prediction accuracy in recommendation systems.
Why It Matters
Continuous value prediction is vital for industries like e-commerce and media streaming, where accurate estimations can significantly impact user engagement and revenue. This research addresses limitations in existing methods, offering a scalable solution that adapts to complex data distributions.
Key Takeaways
- Proposes a residual quantization-based framework for continuous value prediction.
- Addresses challenges of existing generative approaches in modeling complex data distributions.
- Demonstrates improved prediction accuracy through recursive quantization code predictions.
- Outperforms state-of-the-art methods in extensive evaluations.
- Shows strong generalization across various continuous value prediction tasks.
Computer Science > Information Retrieval arXiv:2602.23012 (cs) [Submitted on 26 Feb 2026] Title:Sequential Regression for Continuous Value Prediction using Residual Quantization Authors:Runpeng Cui, Zhipeng Sun, Chi Lu, Peng Jiang View a PDF of the paper titled Sequential Regression for Continuous Value Prediction using Residual Quantization, by Runpeng Cui and 2 other authors View PDF HTML (experimental) Abstract:Continuous value prediction plays a crucial role in industrial-scale recommendation systems, including tasks such as predicting users' watch-time and estimating the gross merchandise value (GMV) in e-commerce transactions. However, it remains challenging due to the highly complex and long-tailed nature of the data distributions. Existing generative approaches rely on rigid parametric distribution assumptions, which fundamentally limits their performance when such assumptions misalign with real-world data. Overly simplified forms cannot adequately model real-world complexities, while more intricate assumptions often suffer from poor scalability and generalization. To address these challenges, we propose a residual quantization (RQ)-based sequence learning framework that represents target continuous values as a sum of ordered quantization codes, predicted recursively from coarse to fine granularity with diminishing quantization errors. We introduce a representation learning objective that aligns RQ code embedding space with the ordinal structure of target values, a...
