Binary Cumulative Encoding meets Time Series Forecasting

2025-05-30

AI Summary

Existing time-series forecasting methods often discretize continuous targets into one-hot encoded categories to improve training stability, yet this approach disregards the inherent ordinal structure and metric relationships among numerical values. To address this, we propose Binary Cumulative Encoding (BCE), the first encoding scheme that maps continuous values to monotonic binary vectors, explicitly preserving ordinality and relative distances. Based on BCE, we design a lightweight CNN architecture incorporating residual connections and dilated convolutions to align with BCE’s hierarchical prediction property. Within a classification-based regression framework, we employ a probabilistic forecasting loss for end-to-end optimization. Experiments across multiple benchmark datasets demonstrate that our method outperforms state-of-the-art models in both point and quantile forecasting accuracy, while reducing parameter count by 30% and accelerating training by 2.1×.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Recent studies in time series forecasting have explored formulating regression via classification task. By discretizing the continuous target space into bins and predicting over a fixed set of classes, these approaches benefit from stable training, robust uncertainty modeling, and compatibility with modern deep learning architectures. However, most existing methods rely on one-hot encoding that ignores the inherent ordinal structure of the underlying values. As a result, they fail to provide information about the relative distance between predicted and true values during training. In this paper, we propose to address this limitation by introducing binary cumulative encoding (BCE), that represents scalar targets into monotonic binary vectors. This encoding implicitly preserves order and magnitude information, allowing the model to learn distance-aware representations while still operating within a classification framework. We propose a convolutional neural network architecture specifically designed for BCE, incorporating residual and dilated convolutions to enable fast and expressive temporal modeling. Through extensive experiments on benchmark forecasting datasets, we show that our approach outperforms widely used methods in both point and probabilistic forecasting, while requiring fewer parameters and enabling faster training.

Problem

• Addresses ordinal structure neglect in one-hot encoding
• Introduces binary cumulative encoding for distance-aware learning
• Proposes efficient CNN architecture for time series forecasting

Innovation

• Binary cumulative encoding preserves ordinal structure
• Convolutional neural network with residual connections
• Distance-aware learning within classification framework