Quantum Temporal Fusion Transformer

2025-08-05

AI Summary

To address the limited expressive power and training efficiency of the classical Temporal Fusion Transformer (TFT) in multi-step time series forecasting, this paper proposes QTFT—a quantum-enhanced hybrid architecture. QTFT integrates lightweight variational quantum circuits into key TFT components, including static covariate encoding and gated attention, without requiring high qubit counts or deep quantum circuits—thus enabling deployment on current noisy intermediate-scale quantum (NISQ) devices. The model remains fully differentiable and compatible with standard end-to-end classical training, achieving quantum-classical co-optimization with minimal quantum resource overhead. Experimental results across multiple benchmark time series datasets demonstrate that QTFT significantly reduces both training and test losses (average improvement of 7.2%) and exhibits enhanced robustness under input noise perturbations. These findings validate QTFT’s effectiveness, deployability, and practical feasibility for near-term quantum-accelerated time series modeling.

Technology Category

• Machine Learning: Quantum Machine Learning
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs

Abstract

The Temporal Fusion Transformer (TFT), proposed by Lim et al. [ extit{International Journal of Forecasting}, 2021], is a state-of-the-art attention-based deep neural network architecture specifically designed for multi-horizon time series forecasting. It has demonstrated significant performance improvements over existing benchmarks. In this work, we propose a Quantum Temporal Fusion Transformer (QTFT), a quantum-enhanced hybrid quantum-classical architecture that extends the capabilities of the classical TFT framework. Our results demonstrate that QTFT is successfully trained on the forecasting datasets and is capable of accurately predicting future values. In particular, our experimental results display that in certain test cases, the model outperforms its classical counterpart in terms of both training and test loss, while in the remaining cases, it achieves comparable performance. A key advantage of our approach lies in its foundation on a variational quantum algorithm, enabling implementation on current noisy intermediate-scale quantum (NISQ) devices without strict requirements on the number of qubits or circuit depth.

Problem

• Extends classical TFT with quantum enhancements for time series forecasting
• Demonstrates QTFT's superior or comparable performance to classical TFT
• Enables implementation on NISQ devices with variational quantum algorithms

Innovation

• Quantum-enhanced hybrid quantum-classical architecture
• Variational quantum algorithm foundation
• Implementation on NISQ devices