Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2609.14529

Parameter-Efficient Quantum NLP for Paraphrase Detection: Performance, Robustness, and Entanglement

Farha Nausheen, Khandakar Ahmed, Farina Riaz

quant-phcs.CLSubmitted Sep 13, 2026• Announced Sep 15, 2026• 0 citations
arXivHTMLPDF

Abstract

Rigorous empirical validation of quantum machine learning on natural language tasks remains scarce. We evaluate a 10-qubit hybrid quantum-classical variational circuit (2,148 parameters) for paraphrase detection across three benchmarks: MRPC, Quora Question Pairs (QQP), and adversarial PAWS. On QQP (n = 10 seeds), the circuit achieves 75.53% +/- 0.75%. accuracy, statistically outperforming parameter-matched classical baselines (DeepMLP: p = 0.015, Cohen's d = 1.20; F1: p < 0.001, d = 2.32) and surpassing DistilBERT-4bit with 31,191 fewer parameters. On MRPC the optimal 2-layer variant reaches 92% of BERT-base accuracy at 54,000 lower parameter cost. Circuit depth analysis reveals a dataset-depth scaling effect; entanglement analysis via the Meyer-Wallach measure identifies multi-qubit entanglement as the primary performance driver (r = 0.85 across four variants). Adversarial evaluation on PAWS reveals emergent robustness: 98.2% recall versus 81.6% classical (+16.6 pp, d = 1.24, p < 0.001), without adversarial training. These results constitute the first systematic parameter-matched multi-benchmark empirical validation of hybrid variational circuits for NLP. All results are from classical simulation; hardware validation is future work.

Related papers

Two Methodological Pitfalls in Matched-Parameter Quantum–Classical Benchmarking for Paraphrase Identification: Class-Imbalance Collapse and Readout-Offset Asymmetry

Nazmus Salehin Sammo, Sariya Akhter Lura, Md Jannatul Rakib Joy +2 more

Quantum machine learning is commonly benchmarked by matching a quantum head’s parameter count to that of a classical baseline. We show that this control is insufficient. We compare a 32-parameter parameterized quantum circuit (PQC) head against a nominally matched 31-parameter multi-layer perceptron (MLP) on three paraphrase tasks—the Microsoft Research Paraphrase Corpus (MRPC), Paraphrase Adversaries from Word Scrambling (PAWS), and Quora Question Pairs (QQP)—and document two pitfalls that inflate an apparent quantum deficit. First, under the tested protocol, unweighted binary cross-entropy on imbalanced data leads to near-all-positive prediction, producing an illusory equivalence. Second, the surviving gap traces to a readout that sits offset from the decision threshold at initialization (⟨Z0⟩=+0.922), and no circuit parameter acts as a threshold shift that could recenter it. A pre-specified calibration experiment removes 61% to 76% of the gap, and a training-free probe detects the condition in one forward pass. Threshold-free metrics then changed our own conclusions. On PAWS, both heads rank barely above chance (area under the receiver operating characteristic curve, AUC, at most 0.508), so the difference there is threshold placement. Only QQP retains a residual, and a post hoc control doubling the qubit count narrows it by 42% without removing it (paired p=0.0009). This pattern is consistent with a capacity or optimization limitation, but does not distinguish between them. Four of five pre-specified predictions failed. Parameter roles and readout offsets must be audited, not only counts.

Electronics•Oct 14, 2026•0 citations

A Hybrid Quantum-Classical Fusion for Deep Semantic Paraphrase Detection

Devanarayanan K, Fayas S Mohamad, Dheeraj V Mohan +1 more

Paraphrase Detection is a core task in natural language processing (NLP) that aims to determine whether two sentences convey equivalent meanings.This work proposes a hybrid quantum-classical framework that integrates Sentence-BERT embeddings, simulated quantum feature encoding, and classical machine learning models to enhance semantic similarity detection.Initially, sentence pairs are embedded using Sentence-BERT and standardized through feature scaling.These representations are then transformed via rotation-based quantum circuits to capture higher-order feature interactions and non-linear dependencies.The resulting hybrid feature space, combining classical and quantum-inspired components are evaluated using LightGBM and deep neural network classifiers.Experimental results shows that the hybrid model incorporating quantuminspired features achieved superior classification performance, yielding a 10% improvement in overall accuracy outperforming standalone deep learning baselines.These findings demonstrate that quantum-classical fusion enhances semantic feature extraction and significantly improves paraphrase detection performance.

•Jan 1, 2025•0 citations

Quantum-enhanced phonological rhyme detection in poetry: A parameter-efficient hybrid CRNN-VQC model

Alireza Rezaei, Azam Bastanfard, Sasan Harifi

Abstract Detecting rhyme in Persian poetry is computationally difficult because the language's written form often diverges from its phonological realization. While massive deep learning models can address this ambiguity, they require excessive computational resources. To solve this efficiency bottleneck, we propose CRNN-VQC, a hybrid architecture that integrates Convolutional-Recurrent Neural Networks (CNN–BiGRU) with Variational Quantum Circuits (VQCs). Instead of standard dense layers, our approach embeds character-level features into an 8-qubit parameterized quantum circuit via a multi-stage data re-uploading scheme, enabling expressive non-linear mapping in a high-dimensional quantum space. We evaluated the model on a hand-curated corpus of 20,171 hemistich pairs, achieving a test accuracy of 96.95% and an F1-score of 96.30%, which statistically outperforms equivalent classical baselines (p < 0.001). Crucially, CRNN-VQC matches the performance of large pre-trained language models like ParsBERT while using approximately three orders of magnitude fewer parameters (163k versus 162M). Furthermore, robustness tests under simulated quantum noise confirm the model’s stability, indicating its feasibility for deployment on Noisy Intermediate-Scale Quantum (NISQ) devices. These findings demonstrate that hybrid quantum–classical frameworks offer a highly parameter-efficient alternative for analyzing the phonology of morphologically rich languages.

•Mar 12, 2026•0 citations

A quantum-classical hybrid neural network for hate speech detection in Spanish

Francisco José Orts, Laura María Donaire, Gloria Ortega +1 more

• Hybrid quantum-classical model for Spanish hate speech detection. • Two-phase training stabilizes quantum circuit optimization. • Competitive with transformers; best results on HaterNet dataset. • Consistently outperforms classical and recurrent baselines. • Demonstrates viability of quantum NLP in real-world tasks. Hate speech detection in social media remains a pressing challenge in natural language processing, particularly for languages such as Spanish where annotated resources are limited. This work proposes a hybrid quantum-classical neural architecture that combines bidirectional gated recurrent units with attention and a variational quantum circuit used as a non-linear classifier. The model is trained in two phases: first the recurrent and attention-based layers are optimized to produce stable representations, then these are frozen and a quantum circuit is fine-tuned for classification. Evaluation on two benchmark corpora, HatEval and HaterNet, shows that the proposed hybrid approach achieves competitive performance with strong transformer baselines such as BETO and XLM-R, while consistently outperforming traditional machine learning and recurrent neural models. On HaterNet, the proposed model performs on par with, and in some metrics slightly better than, the transformer baselines, whereas on HatEval it attains slightly lower scores. Its strength lies in detecting hate speech under class imbalance, as reflected in solid F1 scores for the hate speech class. These findings provide an initial empirical assessment of quantum-enhanced NLP in a realistic hate speech detection scenario and suggest promising directions for further study as quantum hardware matures, without constituting evidence of quantum advantage.

Expert Systems with Applications•Dec 19, 2025•3 citations

Do Quantum Transformers Help? A Systematic VQC Architecture Comparison on Tabular Benchmarks

Chi-Sheng Chen, En-Jui Kuo

Variational quantum circuits (VQCs) are a leading approach to quantum machine learning on near-term devices, yet it remains unclear which circuit architecture yields the best accuracy-parameter trade-off on classical tabular data. We present a systematic empirical comparison of four VQC families -- multi-layer fully-connected (FC-VQC), residual (ResNet-VQC), hybrid quantum-classical transformer (QT), and fully quantum transformer (FQT) -- across five regression and classification benchmarks. Our key findings are: \textbf{(i)}~FC-VQCs achieve 90-96\% of the $R^2$ of attention-based VQCs while using 40-50\% fewer parameters, and consistently outperform equal-capacity MLPs (mean $R^2{=}0.829$ vs.\ MLP$_{720}$'s $0.753$ on Boston Housing, 3-seed average); \textbf{(ii)}~FC-VQC's Type~4 inter-block connectivity provides partial cross-token mixing that approximates the role of attention -- explicit quantum self-attention yields only marginal gains on most datasets while significantly increasing parameter count; \textbf{(iii)}~expressibility saturates at circuit depth~${\approx}\,3$, explaining why shallow VQCs already cover the Hilbert space effectively; \textbf{(iv)}~LayerNorm on the fully quantum transformer improves classification accuracy, suggesting normalization is important when all operations are quantum; \textbf{(v)}~in our noise study on Boston Housing, FQT degrades gracefully under depolarizing noise while QT collapses. All results are validated across three random seeds. These findings provide practical architectural guidance for deploying VQCs on near-term quantum hardware.

quant-pharXiv:2604.23931•Apr 28, 2026•2 citations

Hybrid Quantum-Classical Layers for Sentiment Analysis in Natural Language Processing

Medisetti Esther Rani, Tupakula Sai Sarnya

NLP models have shown remarkable performance for sentiment analysis and text classification tasks due to the deep neural networks and transformer architectures. However, these models typically necessitate considerable computational resources, high parameter complexity and large scale training datasets. Recently, quantum computing has emerged as a promising paradigm that can enhance machine learning systems using quantum parallelism, superposition, and entanglement. In this paper, we propose a new Hybrid Quantum-Classical NLP architecture that combines variational quantum circuit (VQC) layers with classical neural network layers for effective sentiment analysis. The proposed framework combines classical embedding and LSTM layers with a quantum-enhanced variational layer to improve feature representation and classification performance. The proposed framework combines classical embedding and LSTM layers with a quantum-enhanced variational layer to enhance the feature representation and classification performance. The quantum layer is realized via parameterized quantum circuits and optimized using hybrid gradient-based learning techniques. The effectiveness of The proposed framework combines classical embedding and LSTM layers with a quantum-enhanced variational layer to enhance the feature representation and classification performance. The quantum layer is realized via parameterized quantum circuits and optimized using hybrid gradient-based learning techniques. The effectiveness of the proposed model is evaluated on the benchmark sentiment analysis datasets such as IMDb and SST-2 experiments. The hybrid architecture is compared with standard deep learning models in terms of accuracy, training complexity and parameter efficiency. The experimental results show that the proposed quantum-enhanced model can achieve competitive classification accuracy while reducing model complexity and improving contextual semantic learning.

International Conference on Natural Language Processing and Computer Vision (ICNPCV)•Jul 1, 2026•0 citations
Rigorous empirical validation of quantum machine learning on natural language tasks remains scarce. We evaluate a 10-qubit hybrid quantum-classical variational circuit (2,148 parameters) for paraphrase detection across three benchmarks: MRPC, Quora Question Pairs (QQP), and adversarial PAWS. On QQP (n = 10 seeds), the circuit achieves 75.53% +/- 0.75%. accuracy, statistically outperforming parameter-matched classical baselines (DeepMLP: p = 0.015, Cohen's d = 1.20; F1: p < 0.001, d = 2.32) and surpassing DistilBERT-4bit with 31,191 fewer parameters. On MRPC the optimal 2-layer variant reaches 92% of BERT-base accuracy at 54,000 lower parameter cost. Circuit depth analysis reveals a dataset-depth scaling effect; entanglement analysis via the Meyer-Wallach measure identifies multi-qubit entanglement as the primary performance driver (r = 0.85 across four variants). Adversarial evaluation on PAWS reveals emergent robustness: 98.2% recall versus 81.6% classical (+16.6 pp, d = 1.24, p < 0.001), without adversarial training. These results constitute the first systematic parameter-matched multi-benchmark empirical validation of hybrid variational circuits for NLP. All results are from classical simulation; hardware validation is future work.
Quantum machine learning is commonly benchmarked by matching a quantum head’s parameter count to that of a classical baseline. We show that this control is insufficient. We compare a 32-parameter parameterized quantum circuit (PQC) head against a nominally matched 31-parameter multi-layer perceptron (MLP) on three paraphrase tasks—the Microsoft Research Paraphrase Corpus (MRPC), Paraphrase Adversaries from Word Scrambling (PAWS), and Quora Question Pairs (QQP)—and document two pitfalls that inflate an apparent quantum deficit. First, under the tested protocol, unweighted binary cross-entropy on imbalanced data leads to near-all-positive prediction, producing an illusory equivalence. Second, the surviving gap traces to a readout that sits offset from the decision threshold at initialization (⟨Z0⟩=+0.922), and no circuit parameter acts as a threshold shift that could recenter it. A pre-specified calibration experiment removes 61% to 76% of the gap, and a training-free probe detects the condition in one forward pass. Threshold-free metrics then changed our own conclusions. On PAWS, both heads rank barely above chance (area under the receiver operating characteristic curve, AUC, at most 0.508), so the difference there is threshold placement. Only QQP retains a residual, and a post hoc control doubling the qubit count narrows it by 42% without removing it (paired p=0.0009). This pattern is consistent with a capacity or optimization limitation, but does not distinguish between them. Four of five pre-specified predictions failed. Parameter roles and readout offsets must be audited, not only counts.
Paraphrase Detection is a core task in natural language processing (NLP) that aims to determine whether two sentences convey equivalent meanings.This work proposes a hybrid quantum-classical framework that integrates Sentence-BERT embeddings, simulated quantum feature encoding, and classical machine learning models to enhance semantic similarity detection.Initially, sentence pairs are embedded using Sentence-BERT and standardized through feature scaling.These representations are then transformed via rotation-based quantum circuits to capture higher-order feature interactions and non-linear dependencies.The resulting hybrid feature space, combining classical and quantum-inspired components are evaluated using LightGBM and deep neural network classifiers.Experimental results shows that the hybrid model incorporating quantuminspired features achieved superior classification performance, yielding a 10% improvement in overall accuracy outperforming standalone deep learning baselines.These findings demonstrate that quantum-classical fusion enhances semantic feature extraction and significantly improves paraphrase detection performance.
Abstract Detecting rhyme in Persian poetry is computationally difficult because the language's written form often diverges from its phonological realization. While massive deep learning models can address this ambiguity, they require excessive computational resources. To solve this efficiency bottleneck, we propose CRNN-VQC, a hybrid architecture that integrates Convolutional-Recurrent Neural Networks (CNN–BiGRU) with Variational Quantum Circuits (VQCs). Instead of standard dense layers, our approach embeds character-level features into an 8-qubit parameterized quantum circuit via a multi-stage data re-uploading scheme, enabling expressive non-linear mapping in a high-dimensional quantum space. We evaluated the model on a hand-curated corpus of 20,171 hemistich pairs, achieving a test accuracy of 96.95% and an F1-score of 96.30%, which statistically outperforms equivalent classical baselines (p < 0.001). Crucially, CRNN-VQC matches the performance of large pre-trained language models like ParsBERT while using approximately three orders of magnitude fewer parameters (163k versus 162M). Furthermore, robustness tests under simulated quantum noise confirm the model’s stability, indicating its feasibility for deployment on Noisy Intermediate-Scale Quantum (NISQ) devices. These findings demonstrate that hybrid quantum–classical frameworks offer a highly parameter-efficient alternative for analyzing the phonology of morphologically rich languages.
• Hybrid quantum-classical model for Spanish hate speech detection. • Two-phase training stabilizes quantum circuit optimization. • Competitive with transformers; best results on HaterNet dataset. • Consistently outperforms classical and recurrent baselines. • Demonstrates viability of quantum NLP in real-world tasks. Hate speech detection in social media remains a pressing challenge in natural language processing, particularly for languages such as Spanish where annotated resources are limited. This work proposes a hybrid quantum-classical neural architecture that combines bidirectional gated recurrent units with attention and a variational quantum circuit used as a non-linear classifier. The model is trained in two phases: first the recurrent and attention-based layers are optimized to produce stable representations, then these are frozen and a quantum circuit is fine-tuned for classification. Evaluation on two benchmark corpora, HatEval and HaterNet, shows that the proposed hybrid approach achieves competitive performance with strong transformer baselines such as BETO and XLM-R, while consistently outperforming traditional machine learning and recurrent neural models. On HaterNet, the proposed model performs on par with, and in some metrics slightly better than, the transformer baselines, whereas on HatEval it attains slightly lower scores. Its strength lies in detecting hate speech under class imbalance, as reflected in solid F1 scores for the hate speech class. These findings provide an initial empirical assessment of quantum-enhanced NLP in a realistic hate speech detection scenario and suggest promising directions for further study as quantum hardware matures, without constituting evidence of quantum advantage.
Variational quantum circuits (VQCs) are a leading approach to quantum machine learning on near-term devices, yet it remains unclear which circuit architecture yields the best accuracy-parameter trade-off on classical tabular data. We present a systematic empirical comparison of four VQC families -- multi-layer fully-connected (FC-VQC), residual (ResNet-VQC), hybrid quantum-classical transformer (QT), and fully quantum transformer (FQT) -- across five regression and classification benchmarks. Our key findings are: \textbf{(i)}~FC-VQCs achieve 90-96\% of the R2R^2R2 of attention-based VQCs while using 40-50\% fewer parameters, and consistently outperform equal-capacity MLPs (mean R2=0.829R^2{=}0.829R2=0.829 vs.\ MLP720_{720}720​'s 0.7530.7530.753 on Boston Housing, 3-seed average); \textbf{(ii)}~FC-VQC's Type~4 inter-block connectivity provides partial cross-token mixing that approximates the role of attention -- explicit quantum self-attention yields only marginal gains on most datasets while significantly increasing parameter count; \textbf{(iii)}~expressibility saturates at circuit depth~≈ 3{\approx}\,3≈3, explaining why shallow VQCs already cover the Hilbert space effectively; \textbf{(iv)}~LayerNorm on the fully quantum transformer improves classification accuracy, suggesting normalization is important when all operations are quantum; \textbf{(v)}~in our noise study on Boston Housing, FQT degrades gracefully under depolarizing noise while QT collapses. All results are validated across three random seeds. These findings provide practical architectural guidance for deploying VQCs on near-term quantum hardware.
NLP models have shown remarkable performance for sentiment analysis and text classification tasks due to the deep neural networks and transformer architectures. However, these models typically necessitate considerable computational resources, high parameter complexity and large scale training datasets. Recently, quantum computing has emerged as a promising paradigm that can enhance machine learning systems using quantum parallelism, superposition, and entanglement. In this paper, we propose a new Hybrid Quantum-Classical NLP architecture that combines variational quantum circuit (VQC) layers with classical neural network layers for effective sentiment analysis. The proposed framework combines classical embedding and LSTM layers with a quantum-enhanced variational layer to improve feature representation and classification performance. The proposed framework combines classical embedding and LSTM layers with a quantum-enhanced variational layer to enhance the feature representation and classification performance. The quantum layer is realized via parameterized quantum circuits and optimized using hybrid gradient-based learning techniques. The effectiveness of The proposed framework combines classical embedding and LSTM layers with a quantum-enhanced variational layer to enhance the feature representation and classification performance. The quantum layer is realized via parameterized quantum circuits and optimized using hybrid gradient-based learning techniques. The effectiveness of the proposed model is evaluated on the benchmark sentiment analysis datasets such as IMDb and SST-2 experiments. The hybrid architecture is compared with standard deep learning models in terms of accuracy, training complexity and parameter efficiency. The experimental results show that the proposed quantum-enhanced model can achieve competitive classification accuracy while reducing model complexity and improving contextual semantic learning.