Quantum-Inspired Image Encodings for Financial Time-Series Forecasting

ICLR 2026 Conference Submission25645 Authors

Keywords: Time-series Classification, Image Encoding, Quantum Physics, Convolutional Neural Networks (CNN), Financial Forecasting

TL;DR: We propose quantum state–based image encodings for time series that capture both probabilistic amplitudes and dynamic phases, yielding superior forecasting performance over classical methods.

Abstract: This study proposes a quantum-inspired methodology that transforms time-series data into complex-valued image representations for prediction. Unlike classical encodings such as the Gramian Angular Field (GAF), Recurrence Plot (RP), and Markov Transition Field (MTF), which rely on additive pairwise relations, our approach embeds both probabilistic amplitudes and dynamic phase information. Observations are first mapped into quantum amplitudes via Gaussian soft encoding, and local temporal structures are incorporated through phase-function encoding, allowing interference effects that reveal volatility, cumulative imbalances, and phase shifts hidden to classical methods. Building on this foundation, we extend GAF, RP, and MTF into their quantum analogues—Q-GAF, Q-RP, and Q-MTF—producing complex-valued images suitable for CNN-based forecasting. Empirical analysis on the S&P 500 and Russell 3000 indices shows that these quantum-inspired encodings substantially improve predictive accuracy. Our contributions are both methodological and empirical: we present a novel representation framework for financial time series and demonstrate that quantum-inspired image encodings capture richer dynamics and previously undetectable patterns, with implications for forecasting and risk modeling.

Primary Area: other topics in machine learning (i.e., none of the above)

Submission Number: 25645