A Quantum Variational Approach to Prototypical Recurrent Unit

Mahyar Sadeghi Garjan; Tommaso Cesari; Michel Barbeau

cs.LGSep 3, 2026

A Quantum Variational Approach to Prototypical Recurrent Unit

Authors: Mahyar Sadeghi Garjan, Tommaso Cesari, Michel Barbeau

Organizations: School of Electrical Engineering and Computer Science, University of Ottawa, Canada · School of Computer Science, Carleton University, Ottawa, Canada

Abstract

We introduce a lightweight Quantum Prototypical Recurrent Unit (QPRU) that requires significantly fewer parameters than both classical recurrent architectures, such as Long Short- Term Memory (LSTM) and Gated Recurrent Unit (GRU), and quantum variants, including Quantum LSTM (QLSTM) and Quantum GRU (QGRU). Despite its compact design, the QPRU achieves competitive forecasting performance, matching state-of-the-art baselines while offering important structural and practical advantages, including enhanced scalability and a reduced number of trainable parameters.

Figures & tables

Figure 1: Architecture of the proposed QPRU.

Figure 2: VQC exploited in QPRU.

Model	Quantum Params	Total Params
QPRU	$2ql$	$q(2l+d_{\text{in}}+2d_{\text{hid}}+1)+d_{\text{hid}}$
QGRU	$3ql$	$q(3l+2d_{\text{hid}}+d_{\text{in}}+1)+d_{\text{hid}}$
QLSTM	$4ql$	$q(4l+2d_{\text{hid}}+d_{\text{in}}+1)+d_{\text{hid}}$

Table 1: Parameter complexity of QLSTM, QGRU, and the proposed QPRU model.

Model	Classical	Quantum	Total
	Params	Params	Params
QPRU (PRU)	$53\,(80)$	$16\,(0)$	$69\,(80)$
QGRU (GRU)	$53\,(120)$	$24\,(0)$	$77\,(120)$
QLSTM (LSTM)	$53\,(60)$	$32\,(0)$	$85\,(160)$

Table 2: Parameter comparison of classical and quantum recurrent units for hidden state dimension

5

, input feature dimension

1

4

qubits, and

2

ansatz layers. Classical counterparts are shown in parentheses.

Figure 3: Sinusoidal Function Forecasting using QPRU . The red solid line represents the absolute error, and the dashed red line represents the training/validation split.

Model	$\boldsymbol{\ell_{1}}(\times 10^{-2}$ )	$\boldsymbol{\ell_{2}}(\times 10^{-4}$ )	$\boldsymbol{\ell_{\infty}}(\times 10^{-2})$
QPRU	$1.2949\pm 0.4543$	$2.63\pm 1.83$	$2.8277\pm 0.9683$
QGRU	$1.5572\pm 0.6885$	$3.66\pm 2.59$	$3.1338\pm 1.3277$
QLSTM	$1.5772\pm 0.6093$	$4.01\pm 2.74$	$3.7471\pm 1.3887$
PRU	$1.4474\pm 0.5951$	$3.52\pm 3.01$	$3.2643\pm 1.2826$
GRU	$1.7772\pm 0.6611$	$5.10\pm 3.48$	$3.6358\pm 1.2057$
LSTM	$1.5491\pm 0.4759$	$3.36\pm 1.75$	$3.0791\pm 0.8634$

Table 3: Experiment I results. Norm values are averaged over 10 independent validation runs for each model. Values are expressed in

(mean\pm STD)\times 10^{-n}

format.

Figure 4: Forecasting of AAPL closing prices. Predictions on training and validation sets are shown, with the dashed line marking the split. Validation MSE and MAE (computed on denormalized data) are

4.4370\pm 7.920

and

1.6265\pm 1.338

, respectively, indicating consistent performance on unseen data.

Model	$\boldsymbol{\ell_{1}}$	$\boldsymbol{\ell_{2}}$	$\boldsymbol{\ell_{\infty}}$
QPRU	$0.0461\pm 0.0151$	$0.0032\pm 0.0017$	$0.1559\pm 0.0203$
QGRU	$0.1086\pm 0.1017$	$0.0462\pm 0.1082$	$0.3472\pm 0.3590$
QLSTM	$0.1198\pm 0.0906$	$0.0552\pm 0.0837$	$0.5473\pm 0.5261$
PRU	$0.0667\pm 0.0194$	$0.0064\pm 0.0031$	$0.1768\pm 0.0292$
GRU	$0.0606\pm 0.0222$	$0.0052\pm 0.0034$	$0.1573\pm 0.0283$
LSTM	$0.0553\pm 0.0047$	$0.0065\pm 0.0295$	$0.1816\pm 0.1271$

Table 4: Experiment II results. Norm values are averaged over 10 independent validation runs for each model and are computed on data normalized to the range

[-1,1]

. Values of the table are of the form mean

\pm

STD.

Figure 5: Confidence intervals for stock forecasting for different models. The intervals are computed using

r=2\sqrt{\ln(2/\delta)/20}

, with

\delta=0.0005

, representing the half-width of the interval at a high confidence level. The intervals for all methods overlap, indicating that none of the pairwise differences is statistically significant. Within this experimental setup, our QPRU model performs on par with state-of-the-art baselines while retaining the structural and practical advantages discussed earlier.

Explore similar work

Jun 22, 2026quant-ph

Recursive QLSTM with Dynamic Variational Quantum Circuit Adaptation

Recent advances in quantum computing and machine learning have motivated the development of quantum models for sequential data processing. In this paper, we propose a Recursive Quantum Long Short-Term Memory model, or Recursive QLSTM, which extends QLSTM through metacore-based recursive constructions. We numerically test the model under different input sequence lengths, metacore designs, and recursive rules, and identify the best-performing architecture among these variants. For this selected model, we further provide theoretical arguments explaining why its recursive structure improves temporal information propagation and enhances learning performance. Our results suggest that Recursive QLSTM offers a flexible and effective framework for quantum recurrent learning over input time series of various lengths.

Samuel Yen-Chi Chen, Yifeng Peng, Jiun-Cheng Jiang +8

Wells Fargo · Stevens Institute of Technology · National Taiwan University +3

Quantum Machine Learning Recurrent Neural Networks Variational Quantum Circuits Quantum Neural Networks

Ask PDF

Sep 17, 2026cs.LG

Recursive Quantum Long Short-Term Memory for Stable Short-Horizon Temperature Forecasting

Quantum long short-term memory (QLSTM) models extend recurrent sequence learning with variational quantum circuits, but their optimization behavior can vary substantially across random initializations and temporal contexts. This paper evaluates a recursive QLSTM architecture against a standard QLSTM for one-step-ahead prediction of daily minimum and maximum temperature. Using daily weather observations from Toronto and identical training settings, we compare convergence, predictive accuracy, and generalization across input windows of 8, 16, and 32 days over 20 random seeds. The recursive model consistently reaches a near-optimal test loss earlier, reduces mean absolute error and root mean squared error, and exhibits a smaller generalization gap. These results indicate that recursive quantum feature transformations can improve stability and out-of-sample performance for compact hybrid quantum--classical temporal models.

Mu-En Lee, Yen-Ku Liu, Samuel Yen-Chi Chen +1

University of Toronto, Toronto, ON, Canada · National Yang Ming Chiao Tung University, Hsinchu, Taiwan · Brookhaven National Laboratory, Upton, NY, USA +1

Time Series Forecasting Quantum Machine Learning Recurrent Neural Networks Hybrid Quantum-Classical ML Quantum Neural Networks Weather Forecasting

Ask PDF

May 7, 2026cs.LG

Gated QKAN-FWP: Scalable Quantum-inspired Sequence Learning

Fast Weight Programmers (FWPs) encode temporal dependencies through dynamically updated parameters rather than recurrent hidden states. Quantum FWPs (QFWPs) extend this idea with variational quantum circuits (VQCs), but existing implementations rely on multi-qubit architectures that are difficult to scale on noisy intermediate-scale quantum (NISQ) devices and expensive to simulate classically. We propose gated QKAN-FWP, a fast-weight framework that integrates FWP with Quantum-inspired Kolmogorov-Arnold Network (QKAN) using single-qubit data re-uploading circuits as learnable nonlinear activation, known as DatA Re-Uploading ActivatioN (DARUAN). We further introduce a scalar-gated fast-weight update rule that stabilizes parameter evolution, supported by a theoretical analysis of its adaptive memory kernel, geometric boundedness, and parallelizable gradient paths. We evaluate the framework across time-series benchmarks, MiniGrid reinforcement learning, and highlight real-world solar cycle forecasting as our main practical result. In the long-horizon setting with 528-month input window and 132-month forecast horizon, our 12.5k-parameter model achieves lower scaled Mean Square Error (MSE), peak amplitude error, and peak timing error than a suite of classical recurrent baselines with up to 13x more parameters, including Long Short-Term Memory (LSTM) networks (25.9k-89.1k parameters), WaveNet-LSTM (167k), Vanilla recurrent neural network (11.5k), and a Modified Echo State Network (132k). To validate NISQ compatibility, we further deploy the trained fast programmer on IonQ and IBM Quantum processors, recovering forecasting accuracy within 0.1% relative MSE of the noiseless simulator at 1024 shots. These results position gated QKAN-FWP as a scalable, parameter-efficient, and NISQ-compatible approach to quantum-inspired sequence modeling.

Kuo-Chung Peng, Samuel Yen-Chi Chen, Jiun-Cheng Jiang +16

Department of Physics and Center for Theoretical Physics, National Taiwan University, Taipei, Taiwan · National Center for High-Performance Computing, National Institutes of Applied Research, Hsinchu, Taiwan · Wells Fargo, New York, NY, USA +11

Time Series Forecasting Quantum-Inspired ML Kolmogorov-Arnold Networks Variational Quantum Circuits

Ask PDF