Pith. sign in

REVIEW 6 major objections 7 minor 39 references

Multilingual Machine Translation with Quantum Encoder Decoder Attention-based Convolutional Variational Circuits

T0 review · 6 major / 7 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that a 550-parameter quantum encoder-decoder, QEDACVC, performs multilingual translation on OPUS data at accuracy levels comparable to classical transformer models.

desk verdict A novel quantum encoder-decoder architecture buried under an unsound evaluation: the headline accuracy rests on a misdefined metric and implausible BLEU scores. read the letter →

arxiv 2505.09407 v1 pith:PZGPSD6U submitted 2025-05-14 cs.CL cs.AIcs.ET

Subrit Dikshit , Ritu Tiwari , Priyank Jain This is my paper · ORCID
classification cs.CLcs.AIcs.ET
keywords QuantumComputingMultilingualMachineTranslationEncoder-DecoderAttentionConvolutionVariationalCircuitsOPUSCorpusBLEUScore
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

QEDACVC is proposed as a fully quantum alternative to the classical multilingual translation stack. The paper argues that an encoder-decoder built entirely from parameterized quantum circuits—convolution, pooling, attention, and a variational final layer—can translate among English, German, French, and Hindi using OPUS corpus subsets, with 550 total parameters and an accuracy of 82% plus BLEU scores in the 79–89 range. If this is right, it would show that the expensive attention-based neural machinery behind current translation services can be replaced by a tiny quantum circuit without losing translation quality. The paper also reports ablation results indicating that both the quantum convolution and quantum attention layers contribute to the accuracy gain. The headline result depends on the paper's accuracy definition, set out in Equation (3), which the author uses alongside BLEU as the evaluation protocol.

What carries the argument

The load-bearing object is the 8-qubit quantum circuit stack assembled as an encoder-decoder. Quantum convolutional layers apply pairs of parameterized U3 gates with Ising interactions to extract features; quantum pooling layers reduce dimension by measuring half the qubits; a quantum attention circuit replaces the transformer's scaled dot-product attention; a quantum variational circuit—Hadamard, RY rotations, CNOT shifts, and Z measurement—serves as the decoder's fully connected head. These circuit blocks are the entire model: there is no classical transformer underneath, and the 550 trainable parameters all live in the rotation gates. The argument hangs on these circuits' ability to learn translation patterns from token streams, evaluated by the paper's accuracy formula and BLEU.

What would settle it

Take the trained QEDACVC checkpoints and re-score the same test predictions with standard token-level accuracy (only exact target matches count) and with a standard BLEU implementation against the reference sentences. If the true accuracy is near 20–40% rather than 82%, or the BLEU scores fall below typical neural MT numbers for a 10,000-pair training set, the central claim is falsified. Alternatively, replicate training from the OPUS subsets and check whether any model with 550 parameters reaches BLEU >80; a published small-data translation benchmark at that scale would contradict known data requirements.

Watch

Extended reading notes

Core claim

Central claim: QEDACVC, a quantum encoder-decoder with attention-based convolutional variational circuits, is a viable multilingual machine translation architecture. The model encodes text through four pairs of quantum convolutional and pooling layers on an 8-qubit circuit, applies a quantum attention circuit, and decodes through quantum convolutional, pooling, and attention layers ending in a variational circuit that acts as a fully connected layer. Trained on 10,000 English–German–French–Hindi sentence pairs from OPUS, it is reported to reach 81.8% accuracy and 89.2 BLEU for English, 71.9/80.1 for German, 72.8/79.1 for French, and 58.5/81.9 for Hindi, outperforming GRU and LSTM by large margins and matching BERT, GPT, and T5 in the paper's comparison tables. The paper further claims that with 550 parameters and eight layers, the architecture is orders of magnitude smaller than classical baselines while maintaining comparable accuracy. Ablation runs O1–O5 show accuracy increasing monotonically as quantum convolution and quantum attention layers are added, which the paper reads as evidence that each quantum component carries real modeling weight.

Load-bearing premise

The load-bearing premise is that the paper's accuracy formula, which treats incorrect responses to predicted tokens as true negatives, and its BLEU implementation measure translation quality in the standard way; if either is nonstandard, the headline results collapse.

Editorial extensions

If this is right

  • A quantum encoder-decoder with only 550 parameters can run multilingual translation for four languages at accuracy levels the paper reports as comparable to BERT, GPT, and T5.
  • Adding quantum convolution layers and a quantum attention layer each improves accuracy across all four languages in the ablation study.
  • The architecture's dependence on open quantum simulators and free quantum hardware suggests translation models of this kind can be trained without large compute clusters.
  • Because the loss in Equation (1) blends encoder and decoder token probabilities, the paper implies the quantum circuit stack can be trained end-to-end by gradient-based optimization.
  • The reported metric gap between English (81.8) and Hindi (58.5) is attributed to dataset constraints, implying that larger or better-balanced OPUS subsets would raise lower-resource-language performance.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A BLEU score above 80 on a 10,000-pair training set is far beyond what classical neural MT typically achieves; if the numbers are reproducible, they would suggest the evaluation compares against a very small or memorized test set, or that the BLEU implementation is not the standard reference-based metric.
  • Equation (3)'s true-negative convention likely inflates accuracy substantially: in a vocabulary of tens of thousands of tokens, most positions the model does not predict are 'incorrect responses to predicted tokens,' so a model predicting almost nothing could still score high.
  • A direct test of the quantum components' contribution would be to replace each quantum layer with a random or identity circuit; if accuracy stays high, the ablations measure parameter count rather than circuit structure.
  • If the framework scales, the natural next experiment is to compare QEDACVC against a classical transformer of similar parameter count on the same 10,000-pair split, which the paper does not report.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

6 major / 7 minor

Summary. The paper proposes QEDACVC, a quantum encoder-decoder architecture that combines quantum convolutional, pooling, attention, and variational circuit components, and applies it to multilingual machine translation on English, German, French, and Hindi subsets of the OPUS corpus. The authors report 82% accuracy for English and BLEU scores of 89.2 (English), 80.1 (German), 79.1 (French), and 81.9 (Hindi) with a model of 550 total parameters, and compare these results against GRU, LSTM, BERT, GPT, and T5 baselines. They also present an ablation study that varies quantum convolution and attention components. The central claim is that QEDACVC performs comparatively well against classical state-of-the-art models while using far fewer parameters.

Significance. If the reported results were valid, a 550-parameter quantum model achieving near-state-of-the-art multilingual translation would be a striking contribution to quantum natural language processing, especially given the scarcity of work on quantum machine translation. The paper also demonstrates a genuine attempt to assemble quantum circuit components for encoder-decoder translation and includes an ablation study that identifies the contribution of individual components. However, the evaluation is deeply flawed: the accuracy metric is nonstandard to the point of being meaningless, the BLEU scores are implausible, the experimental configuration is internally inconsistent, and the baselines are not described. These issues are load-bearing, because the paper's headline accuracy and all comparative conclusions rest on this evaluation. The manuscript would need a complete re-run of experiments with standard metrics, reproducible code, and a rigorously described baseline protocol before its claims could be assessed.

major comments (6)
  1. [§4, Eq. (3)] The accuracy metric in Eq. (3) is invalid. The paper defines true negative (β) as "Number of incorrect responses to predicted tokens"; in standard classification, a true negative is a correct rejection, not an error. Counting model errors as successes inflates the accuracy numerator by the model's own mistakes and makes higher accuracy partially a reward for incorrect outputs. Moreover, sequence-to-sequence translation has no well-defined set of negative tokens at each output position, so the four-way confusion matrix cannot be applied meaningfully. Because Table 3 uses these accuracies to compare QEDACVC against classical baselines without stating that the baselines used the identical nonstandard definition, all accuracy comparisons in the paper are invalid. This is the core evaluation claim and cannot be fixed by textual revision; the experiments must be re-run with a standard metric.
  2. [§4, Table 3 and Fig. 11] The reported BLEU scores—89.2 for English, 80.1 for German, 79.1 for French, and 81.9 for Hindi—are far above typical results in machine translation, especially for a model trained on only 10,000 sentence pairs and containing only 550 parameters. The paper provides no code, no exact data split, no tokenization details, no beam-search width or decoding strategy, and no description of the BLEU implementation used. Without these, the BLEU numbers are unreproducible and not credible. The conclusion that QEDACVC "performs comparatively well" against BERT, GPT, and T5 depends entirely on these unverifiable scores.
  3. [§3, Algorithm 1, Step 5 vs. §4, Training] The hyperparameters are stated inconsistently. Algorithm 1, Step 5 specifies dropout rate 0.02, epochs 100, learning rate 1e-5, batch size 8, and sequence length 64; Section 4, Training states dropout rate 0.02, epochs 50, learning rate 2e-7, batch size 8, and maximum sequence length 64. These discrepancies mean that the reported results are not tied to a single, reproducible configuration and undermine the validity of every training curve and evaluation number in the paper.
  4. [Table 2] The validation-loss column contains repeated identical values across different language configurations at the same epoch (for example, 0.215 at epoch 10 for English and Hindi, 0.167 at epoch 15, and 0.157 at epoch 50). Such exact repetitions are implausible for independent training runs on different languages and suggest that the reported values may be corrupted or copied. Additionally, the training narrative claims stabilization around epochs 32, 32, 43, and 35 for English, German, French, and Hindi, respectively, but Table 2 lists only epochs in increments of five, so these claimed stabilization points do not correspond to any logged row, making the training description internally inconsistent.
  5. [§4, baselines] The paper does not describe how the GRU, LSTM, BERT, GPT, and T5 baselines were implemented, trained, or evaluated. There is no information about their data splits, hyperparameters, training setup, tokenization, decoding, or which implementation of the accuracy and BLEU metrics was used. Without this information, the comparison in Table 3 and Fig. 11 is not meaningful; it is unclear whether the baselines were evaluated under conditions even roughly comparable to QEDACVC or whether they used the same nonstandard accuracy definition.
  6. [Abstract and §1] The abstract and the highlights claim an accuracy of 82%, but Table 3 shows that this value applies only to English; the corresponding accuracies are 71.9% for German, 72.8% for French, and 58.5% for Hindi. Reporting only the best language as the headline 'multilingual translation' accuracy is misleading and does not represent a multilingual average. The paper should state per-language results and an aggregate measure if one is intended.
minor comments (7)
  1. [§2 and §8] Reference [22] is cited in the text as "Myles Doyle et al. (2020)" but the reference list entry for [22] is "Mina A, Vahid S, Seyed M, et al., Application of Quantum Natural Language Processing for Language Translation," which is an Abbaszade et al. work with different authors and year. The citation is mismatched and needs correction.
  2. [§2, reference [24]] The text dates Google Neural Machine Translation System as (2022), but the reference list entry is the 2016 arXiv paper (arXiv:1609.08144). The year should be corrected.
  3. [§4, Fig. 10] The sample inferences in Fig. 10(a-c) are labeled with language pairs, but the figure does not show source sentences, reference translations, or the target-language output in a way that allows assessment. The captions should make clear what is being translated and what the expected output is.
  4. [§4, Eq. (2) and surrounding text] Equation (2) defines attention with "M" as the softmax activation, but the description says "M, A is the ask, B is the key, and C is the value," where "ask" appears to be a typo for "query." The notation should be clarified and the typo fixed.
  5. [§5, Tables 5 and 6] Table 5 is titled "Model Vs. Parameter-Based Performance Impact" and the column is called "Parameter Optimization Mode," but the rows O1-O5 describe ablated model configurations, not parameters being optimized. Renaming the column to "Configuration" would better match the content.
  6. [§4, Training] The sentence "Training QEDACVC is trained for a zero-shot multilingual setup" is unclear, because the model is explicitly trained on all four language pairs. The use of "zero-shot" does not match the standard terminology for multilingual machine translation.
  7. [Throughout] No error bars, standard deviations, or repeated-run statistics are reported for any accuracy or BLEU number. Given the small batch size (8) and limited training data, reporting variance across seeds is necessary to support the claimed comparisons.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity; the sole self-citation is background-only, and the reported metrics, while methodologically questionable, are not derived from the model's own assumptions.

full rationale

The paper's central claim—that QEDACVC achieves 82% accuracy and high BLEU scores on OPUS subsets—is an empirical, experimentally reported result, not a quantity derived from the model's own definitions or fitted parameters. The loss in Eq. (1), the attention formula in Eq. (2), and the accuracy definition in Eq. (3) are measurement or training definitions, not predictions that reduce to their inputs by construction. Eq. (3) mislabels 'incorrect responses to predicted tokens' as true negatives, which is a serious evaluation-validity flaw, but it is not circularity: the reported accuracy still depends on empirical counts and is not equivalent to an assumed input. The only self-citation is reference [1], used in the introduction to cite 'modern multilingual state-of-the-art networks' as background; it is not load-bearing for the architecture, the training, or the results. The quantum components (convolution, pooling, variational circuits, attention) are adopted from external cited works and are not used to smuggle in the paper's conclusion. Concerns about the nonstandard accuracy metric, implausibly high BLEU scores, and missing code or decoding details belong to correctness and reproducibility review, not circularity analysis. Accordingly, no step identified satisfies the standard for circularity, and the score of 1 reflects only the presence of one minor, non-load-bearing self-citation.

Assumptions & free parameters 8 free parameters · 4 assumptions · 0 invented entities

The central results are empirical; the paper introduces no new physical entities. It does rely on many hand-chosen hyperparameters and unvalidated modeling assumptions, which are listed.

free parameters (8)
  • learning_rate = 1e-5 or 2e-7 (conflicting)
    Set in Step 5 of Algorithm 1 as 1e-5, but Section 4 Training states 2e-7; no justification or sensitivity analysis.
  • dropout_rate = 0.02
    Chosen without ablation, though four other parameter configurations are tested.
  • epochs = 100 (Step 5) or 50 (Section 4)
    Conflicting values; training curves go to 50 epochs.
  • batch_size = 8
    Set in both descriptions; arbitrary.
  • max_sequence_length = 64
    Used for tokenization padding and truncation.
  • number_of_qubits = 8
    Quantum circuits initialized for 8 qubits; no justification.
  • quantum_convolution_pooling_layers = 4
    Encoder uses four four-layer quantum convolutional and pooling blocks.
  • dataset_split_sizes = train 10000, test 3000, validation 1000
    Randomized splits chosen without specifying how the OPUS subsets were selected.
assumptions (4)
  • domain assumption Standard quantum circuit model (qubits, gates, measurements) is used.
    The architecture depends on the realizability of the described circuits on quantum hardware or simulators.
  • ad hoc to paper The loss function in Eq. (1) is a valid cross-entropy-style objective for sequence generation.
    The formula multiplies encoder and decoder probabilities and is not a standard cross-entropy; training success hinges on this objective.
  • ad hoc to paper The accuracy metric in Eq. (3) measures translation quality.
    True negatives are defined as 'incorrect responses to predicted tokens', which is nonstandard and likely inflates accuracy.
  • domain assumption Baseline models (GRU, LSTM, BERT, GPT, T5) are trained under comparable conditions.
    No details are given on how baselines are trained, so the comparison depends on an unverified fairness assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multilingual Machine Translation with Quantum Encoder Decoder Attention-based Convolutional Variational Circuits." pith.science (2026). https://pith.science/paper/PZGPSD6U

@misc{pith2026250509407,
  author       = {Pith},
  title        = {Pith review of: Multilingual Machine Translation with Quantum Encoder Decoder Attention-based Convolutional Variational Circuits},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PZGPSD6U}},
  note         = {Machine review of arXiv:2505.09407}
}
read the original abstract

Cloud-based multilingual translation services like Google Translate and Microsoft Translator achieve state-of-the-art translation capabilities. These services inherently use large multilingual language models such as GRU, LSTM, BERT, GPT, T5, or similar encoder-decoder architectures with attention mechanisms as the backbone. Also, new age natural language systems, for instance ChatGPT and DeepSeek, have established huge potential in multiple tasks in natural language processing. At the same time, they also possess outstanding multilingual translation capabilities. However, these models use the classical computing realm as a backend. QEDACVC (Quantum Encoder Decoder Attention-based Convolutional Variational Circuits) is an alternate solution that explores the quantum computing realm instead of the classical computing realm to study and demonstrate multilingual machine translation. QEDACVC introduces the quantum encoder-decoder architecture that simulates and runs on quantum computing hardware via quantum convolution, quantum pooling, quantum variational circuit, and quantum attention as software alterations. QEDACVC achieves an Accuracy of 82% when trained on the OPUS dataset for English, French, German, and Hindi corpora for multilingual translations.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 22 canonical work pages

  1. [1]

    In the 2000s, artificial intelligence and deep learning -based systems became prevalent and took over the world by storm

    Introduction In the 1940s, electronics-based computing became widespread, while the 1970s were the era of microprocessor-based classical computing. In the 2000s, artificial intelligence and deep learning -based systems became prevalent and took over the world by storm . Many modern multilingual state -of-the-art [1] networks and cloud-based translation se...

    work page

  2. [2]

    R Narayan et al

    Related Works This section debates the associated efforts of scholars, academicians, and works that are comparable to QEDACVC and its recital. R Narayan et al. (2014) [19] present a machine translation system that utilizes a quantum neural network (QNN) to learn semantic patterns between Hindi and English sentences. The system analyses parts of speech and...

    work page 2014

  3. [3]

    Table 1 Algorithmic definition of the QEDACVC network implementation

    Methodology A formal algorithmic definition of the overall approach and the important philosophies utilized in the QEDACVC network are stated below (Table 1). Table 1 Algorithmic definition of the QEDACVC network implementation. Algorithm 1: Overall Implementation Step 0. Set up Environment and Libraries. Install the following requirement packages: python...

    work page

  4. [4]

    with an arbitrary unitary. Fig. 4. Quantum Dense Circuit. This complete process is depicted in (Fig. 5). The text stream of data is embedded via quantum encoding, next four four- layer quantum convolutional, and quantum pooling is applied. This latent output is condensed with a quantum dense circuit- based layer, and the final output is measured. Fig. 5. ...

    work page

  5. [5]

    Dataset: QEDACVC is trained and fine-tuned on the English, French, German, and Hindi language subsets from the OPUS corpus

    Experiment & Results In this section, the experimental setting and conclusion are discussed. Dataset: QEDACVC is trained and fine-tuned on the English, French, German, and Hindi language subsets from the OPUS corpus. OPUS dataset entails 1212 corpora, 58.851 trillion sentence pairs, and 747 human-spoken. Evaluation Strategy: QEDACVC is evaluated on metric...

    work page

  6. [6]

    Accuracy – 𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 refers to the proportion of truthful estimates when compared with the entire examinations and expressed as: 7 𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = (𝛼 + 𝛽) (𝛼 + 𝛽 + ϒ + δ) (3) • True positive (α): Number of right responses to predicted tokens • True negative (β): Number of incorrect responses to predicted tokens • False positive (ϒ): Number of predicted tokens b...

    work page

  7. [7]

    The BLEU score varies from 0-1, where predictions near 0 values resemblance dissimilarities and values close to 1 indicate similarities

    BLEU Score – 𝐵𝐿𝐸𝑈, referred to as Bilingual assessment, estimates machine-translated texts and reports a stronger connection to judgments made by individuals . The BLEU score varies from 0-1, where predictions near 0 values resemblance dissimilarities and values close to 1 indicate similarities. Training: Training QEDACVC is trained for a zero-shot multil...

    work page

  8. [8]

    Ablation Studies This section performs experiments to observe the parameter -based effect on the performance of different QEDACVC network configurations for English, German, French, and Hindi. The different parameter-based setups are shown in Table 5, where the base model is compared with arran gements, such as the consequence of no quantum convolution la...

    work page

Show all 39 references
  1. [9]

    Conclusion & Future Work This paper studies and discusses an effective multilingual network for the quantum realm that performs well when compared to leading state -of-the-art NLP networks running on humongous machines. The network is not 1:1 completely comparable with leading...

  2. [10]

    Declarations 7.1 Competing interests The corresponding author declares that there is no conflict of interest on the part of all authors. We (the authors) declare that we do not possess any competing interests to reveal, which could be financial or personal relationships with a...

  3. [11]

    Dikshit S, Dixit R, Shukla A ., Review and analysis for state -of-the-art NLP models , International Journal of Systems, Control and Communications 15(1) (2024) 48-78, https://doi.org/10.1504/IJSCC.2024.10060461

    2024

  4. [12]

    Ray PP, ChatGPT: A comprehensive review on background, applications, key challenges, bias, ethics, limitations , and future scope, Internet of Things and Cyber-Physical Systems 3 (2023) 121-154, https://doi.org/10.1016/j.iotcps.2023.04.003

    2023 doi

  5. [13]

    Guo D, Yang D, Zhang H, et al., DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning (2025) arXiv preprint, arXiv:2501.12948

    2025 arXiv

  6. [14]

    Shen G, Tan Q, Zhang H, et al., Deep Learning with Gated Recurrent Unit Networks for Financial Sequence Predictions , Procedia Computer Science 131 (2018) 895-903, https://doi.org/10.1016/j.procs.2018.04.298

    2018 doi

  7. [15]

    Malashin I, Tynchenko V, Gantimurov A , et al., Applications of Long Short -Term Memory (LSTM) Networks in Polymeric Sciences: A Review, Polymers (2024) 2607, https://doi.org/10.3390/polym16182607. 12

    2024 doi

  8. [16]

    Pires T, Schlinger E, and Garrette D. How Multilingual is Multilingual BERT? In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics , Association for Computational Linguistics (2019) 4996–5001, https://doi.org/10.18653/v1/P19-1493

    2019 doi

  9. [17]

    Radford A, Narasimhan K, Salimans T, et al., Improving Language Understanding by Generative Pre-Training (2018) arXiv preprint, arXiv:1801.06146

    2018 arXiv

  10. [18]

    Raffel C, Shazeer N, Roberts A, et al., Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer, The Journal of Machine Learning Research 21(1) (2020) 5485–5551

    2020

  11. [19]

    Vaswani A, Shazeer N, Parmar N , et al., Attention Is All You Need , In Proceedings of the 31st International Conference on Neural Information Processing Systems (NIPS'17), New York, USA (2017) 6000–6010, arXiv:1706.03762

    2017 arXiv

  12. [20]

    Henderson M, Shakya S and Pradhan S, et al., Quanvolutional Neural Networks: Powering Image Recognition with Quantum Circuits (2019) arXiv preprint, arXiv:1904.04767

    2019 arXiv

  13. [21]

    Monnet M, Hanady Gebran H, Flierl AM, et al., Pooling techniques in hybrid quantum -classical convolutional neural networks (2023) arXiv preprint, arXiv:2305.05603

    2023 arXiv

  14. [22]

    Stokes J, Izaac J, Killoran N, et al., Quantum Natural Gradient (2019) arXiv preprint, arXiv:1909.02108

    2019 arXiv

  15. [23]

    Li G, Zhao X, Wang X, Quantum Self -Attention Neural Networks for Text Classification (2022) arXiv preprint , arXiv:2205.05625

    2022 arXiv

  16. [24]

    Baltic Journal of Modern Computing (2016), https://aclanthology.org/2016.eamt-2.8/

    Tiedemann J, OPUS – parallel corpora for everyone , In Proceedings of the 19th Annual Conference of the European Association for Machine Translation: Projects/Products, Riga, Latvia. Baltic Journal of Modern Computing (2016), https://aclanthology.org/2016.eamt-2.8/

    2016

  17. [25]

    , PennyLane: Automatic differentiation of hybrid quantum -classical computations (2018) arXiv preprint, arXiv:1811.04968

    Bergholm V, Izaac J, Schuld M, et al. , PennyLane: Automatic differentiation of hybrid quantum -classical computations (2018) arXiv preprint, arXiv:1811.04968

    2018 arXiv

  18. [26]

    Paszke A, Gross S, Massa F, et al., PyTorch: an imperative style, high-performance deep learning library, Proceedings of the 33rd International Conference on Neural Information Processing Systems , Curran Associates Inc., New York, USA, 721 (2019) 8026–8037

    2019

  19. [27]

    Abadi M, Barham P, Chen J, et al., TensorFlow: a system for large -scale machine learning , In Proceedings of the 12th USENIX conference on Operating Systems Design and Implementation (OSDI'16) USENIX Association, USA, (2016) 265– 283

    2016

  20. [28]

    , JAX: composable transformations of Python and NumPy programs (2018) http://github.com/jax-ml/jax

    Bradbury J, Frostig R, Hawkins P, et al. , JAX: composable transformations of Python and NumPy programs (2018) http://github.com/jax-ml/jax

    2018

  21. [29]

    Narayan R, Chakraverty S, Singh VP, Quantum neural network based machine translator for English to Hindi , Applied Soft Computing, 38 (2016) 1060-1075, https://doi.org/10.1016/j.asoc.2015.08.031

    2016 doi

  22. [30]

    Devlin J, Chang MW, Lee K, et al., BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding (2018), arXiv preprint, arXiv:1810.04805

    2018 arXiv

  23. [31]

    Jiatao G, Awadalla H, Jacob D, Universal Neural Machine Translation for Extremely Low Resource Languages , Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 1 (2018), https://doi.org/...

    2018 doi

  24. [32]

    Mina A, Vahid S, Seyed M, et al., Application of Quantum Natural Language Processing for Language Translation, IEEE Access, 9 (2021) 130434-130448, https://doi.org/10.1109/ACCESS.2021.3108768

    2021

  25. [33]

    Johannes B, Recurrent Quantum Neural Networks, Advances in Neural Information Processing Systems, Curran Associates Inc., 33 (2020) 1368-1379, Recurrent Quantum Neural Networks

    2020

  26. [34]

    Wu Y, Schuster M, Chen Z, et al., Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation (2016) arXiv preprint, arXiv:1609.08144

    2016 arXiv

  27. [35]

    Eriguchi A, Xie S, Qin T, et al., Building Multilingual Machine Translation Systems That Serve Arbitrary X-Y Translations (2022) arXiv preprint, arXiv:2206.14982

    2022 arXiv

  28. [36]

    Shliazhko O, Fenogenova A, Tikhonova M, et al., mGPT: Few-Shot Learners Go Multilingual (2022) arXiv preprint , arXiv:2204.07580

    2022 arXiv

  29. [37]

    Abbaszade M, Zomorodi M, Salari V, et al., Toward Quantum Machine Translation of Syntactically Distinct Languages (2023) arXiv preprint, arXiv:2307.16576

    2023 arXiv

  30. [38]

    Chen P, Ye J, Chen G, et al., Robustness of Accuracy Metric and its Inspirations in Learning with Noisy Labels , AAAI Conference on Artificial Intelligence (2020), https://doi.org/10.1609/aaai.v35i13.17364

    2020 doi

  31. [39]

    Kishore P, Salim R, Todd W, et al., BLEU: a method for automatic evaluation of machine translation , Proceedings of the 40th Annual Meeting on Association for Computational Linguistics, Association for Computational Linguistics, Philadelphia, Pennsylvania, USA, (2002) 311–318,...

    2002

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.