arxiv: 2604.26571 · v1 · submitted 2026-04-29 · 💻 cs.LG · physics.chem-ph· physics.data-an

Recognition: unknown

Advancing multi-site emission control: A physics-informed transfer learning framework with mixture of experts for carbon-pollutant synergy

Yuxuan Ying , Hanqing Yang , Kaige Wang , Yu Hu , Zhiming Zheng , Yunliang Jiang , Xiaoqing Lin , Xiaodong Li

show 1 more author

Jun Chen

Authors on Pith no claims yet

Pith reviewed 2026-05-07 11:27 UTC · model grok-4.3

classification 💻 cs.LG physics.chem-phphysics.data-an

keywords transfer learningmixture of expertsphysics-informed modelingemission predictionwaste incinerationcarbon emissionsair pollutantsmulti-site transfer

0 comments

The pith

A mixture-of-experts model with physics regularization transfers emission predictions across incineration plants while preserving carbon-pollutant coupling.

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

The paper establishes that emission behavior at multiple waste incineration sites can be captured in transferable structures by jointly encoding physical conservation laws, differences in operating regimes, and interactions between carbon and other pollutants. It introduces a framework that routes data through regime-specific experts, applies conservation constraints as penalties during training, and computes a single index for combined emission risks. This matters because plant-specific models lose accuracy when applied elsewhere, limiting efforts to manage emissions at scale across heterogeneous facilities. Tests on 13 plants show retained performance after transfer, with adaptation occurring through re-weighting of known regimes rather than full retraining.

Core claim

Multi-site emission behaviour can be represented through transferable system-level structures when physical constraints, operating-regime heterogeneity and carbon-pollutant coupling are jointly considered. The carbon-pollutant mixture-of-experts model combines regime-dependent expert routing with conservation-based regularization and a carbon-pollutant synergistic index, yielding source-domain pollutant R² of 0.668-0.904 and CPSI R² of 0.666-0.970; after transfer to 12 target plants, pollutant R² stays between 0.661 and 0.842 while CPSI R² ranges from 0.610 to 0.841, with expert patterns showing structured regime re-weighting.

What carries the argument

Carbon-pollutant mixture-of-experts model that routes inputs to regime-specific experts, enforces conservation laws via regularization terms, and evaluates integrated risk through a carbon-pollutant synergistic index.

If this is right

Pollutant predictions remain accurate after transfer with R² values of 0.661-0.842.
CPSI risk scores transfer with comparable R² of 0.610-0.841.
Adaptation to new plants occurs by re-weighting existing operating regimes instead of learning entirely new models.
The representation supports extension into a digital twin for regime-aware operational decisions.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Centralized training on a few reference plants could support emission management for many additional facilities without collecting full datasets at each.
Regime-aware control strategies derived from the model might allow operators to adjust incinerator settings to reduce combined carbon and pollutant loads.
The same joint encoding of physics, regimes, and coupling could be tested on other industrial systems that emit multiple linked pollutants.

Load-bearing premise

Physical conservation constraints and carbon-pollutant coupling can be encoded effectively as regularization terms while mixture-of-experts routing fully captures operating-regime differences across all plants.

What would settle it

Training the model on one plant and testing on a fourteenth plant where the transferred pollutant or CPSI R² drops below the accuracy of a model trained from scratch on the new plant's data alone.

read the original abstract

Municipal solid waste incineration is increasingly central to urban waste management, yet its sustainability benefit depends on controlling carbon emissions and multiple air pollutants under highly heterogeneous operating conditions. Current data-driven models are often accurate within individual plants but are difficult to transfer across facilities, limiting their value for scalable emission-control strategies. Here we show that multi-site emission behaviour can be represented through transferable system-level structures when physical constraints, operating-regime heterogeneity and carbon--pollutant coupling are jointly considered. We develop a physics-informed transfer learning framework built on a carbon--pollutant mixture-of-experts model, which combines regime-dependent expert routing with conservation-based regularization and a carbon--pollutant synergistic index for integrated risk evaluation. Across 13 municipal solid waste incineration plants, the model captured both pollutant-specific emissions and system-level risk, achieving source-domain average pollutant $R^2$ values of 0.668--0.904 and CPSI $R^2$ values of 0.666--0.970. After transfer from a reference facility to 12 target plants, average pollutant $R^2$ remained between 0.661 and 0.842, while CPSI retained comparable transferability ($R^2$ = 0.610--0.841). Expert-utilization patterns further indicate that adaptation occurs through structured re-weighting of operating regimes rather than complete model re-learning. By extending the learned representation into an interpretable digital twin, this framework provides a route from emission prediction to regime-aware operational navigation, supporting scalable carbon--pollutant synergistic control across heterogeneous waste-to-energy systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper applies physics-informed MoE transfer learning plus a new CPSI metric to emission prediction across 13 waste incineration plants and reports usable transfer R², but provides no ablations or constraint checks to confirm the physics and regime components are active.

read the letter

The core contribution is a transfer learning framework that routes through mixture-of-experts for operating regimes, adds conservation regularization, and introduces a carbon-pollutant synergistic index to handle multi-site incinerator data. It shows source R² for pollutants between 0.668 and 0.904, with transfer to 12 target plants holding at 0.661-0.842 and similar numbers for the CPSI. That multi-plant scale and the structured adaptation claim are the parts worth noting; the expert utilization patterns suggest the model is not just retraining from scratch each time.

Referee Report

3 major / 1 minor

Summary. The paper proposes a physics-informed transfer learning framework using a carbon-pollutant mixture-of-experts (MoE) model for emission prediction and control across 13 municipal solid waste incineration plants. It incorporates conservation-based regularization, regime-dependent expert routing, and a newly defined Carbon-Pollutant Synergistic Index (CPSI) to jointly model pollutant emissions and system-level risk. The central claim is that this approach yields transferable performance, with source-domain pollutant R² of 0.668–0.904 and CPSI R² of 0.666–0.970, and post-transfer pollutant R² of 0.661–0.842 and CPSI R² of 0.610–0.841, with adaptation occurring via structured regime re-weighting rather than full re-learning.

Significance. If the transferability and physical constraint enforcement can be rigorously validated, the framework could support scalable, regime-aware emission control strategies and digital twins for heterogeneous waste-to-energy systems. The combination of MoE routing with conservation regularization and the CPSI for integrated risk assessment represents a potentially useful integration of physics-informed methods and multi-task transfer learning in environmental engineering.

major comments (3)

[Abstract/Methods] Abstract and Methods: The reported R² ranges for source and target domains provide no details on validation splits, cross-validation procedure, error bars, or explicit checks for data leakage during transfer from the reference facility to the 12 target plants. These omissions are load-bearing for the transfer claim, as post-hoc regime definitions and potential leakage could inflate the observed performance.
[CPSI definition section] CPSI definition and evaluation: The Carbon-Pollutant Synergistic Index is introduced as an invented entity whose performance is evaluated on quantities derived from the same fitted model. Without explicit independent equations or external grounding (e.g., comparison to measured synergistic effects), the reported CPSI transfer R² values risk circularity and reduce to in-sample fit quality.
[Results/Experiments] Results on mechanisms: No ablation studies, expert-routing visualizations, or constraint-violation metrics (such as mass-balance residuals under varying regularization weights) are provided to demonstrate that conservation regularization and MoE routing actively enforce physical constraints and regime heterogeneity. Aggregate R² alone does not confirm the mechanisms are operative rather than incidental to standard transfer learning.

minor comments (1)

[Methods] Notation for the conservation regularization coefficients and expert routing parameters should be defined more explicitly with equations to improve reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight important aspects of validation, grounding, and mechanistic evidence. We address each point below and will revise the manuscript to incorporate the requested details and analyses.

read point-by-point responses

Referee: [Abstract/Methods] Abstract and Methods: The reported R² ranges for source and target domains provide no details on validation splits, cross-validation procedure, error bars, or explicit checks for data leakage during transfer from the reference facility to the 12 target plants. These omissions are load-bearing for the transfer claim, as post-hoc regime definitions and potential leakage could inflate the observed performance.

Authors: We agree that transparent validation details are essential. In the revised manuscript, we will expand the Methods section to specify the temporal train/validation/test splits used to prevent leakage, the cross-validation procedure for model selection, standard error bars on all reported R² values, and explicit checks confirming no data leakage in the transfer from the reference plant to the 12 targets. These additions will directly bolster the transferability claims. revision: yes
Referee: [CPSI definition section] CPSI definition and evaluation: The Carbon-Pollutant Synergistic Index is introduced as an invented entity whose performance is evaluated on quantities derived from the same fitted model. Without explicit independent equations or external grounding (e.g., comparison to measured synergistic effects), the reported CPSI transfer R² values risk circularity and reduce to in-sample fit quality.

Authors: The CPSI is defined via an independent physics-based coupling equation linking carbon and pollutant emissions, separate from model outputs. To address circularity concerns, the revision will (i) present the explicit standalone CPSI equations and (ii) include comparisons against available plant operational risk indicators where external data permits. This provides additional grounding beyond in-sample fits. revision: partial
Referee: [Results/Experiments] Results on mechanisms: No ablation studies, expert-routing visualizations, or constraint-violation metrics (such as mass-balance residuals under varying regularization weights) are provided to demonstrate that conservation regularization and MoE routing actively enforce physical constraints and regime heterogeneity. Aggregate R² alone does not confirm the mechanisms are operative rather than incidental to standard transfer learning.

Authors: We concur that mechanistic validation strengthens the claims. The revised manuscript will add ablation studies isolating the conservation regularization and MoE routing, visualizations of expert utilization patterns across operating regimes, and quantitative constraint-violation metrics (e.g., mass-balance residuals) evaluated under different regularization weights. These will demonstrate the active contribution of the physics-informed components. revision: yes

Circularity Check

0 steps flagged

No significant circularity; claims rest on empirical transfer evaluation rather than definitional reduction

full rationale

The abstract and description introduce a new CPSI index and report R² metrics on source and transferred domains as performance evidence. No equations or self-citations are provided that would allow the reported transfer R² values or CPSI scores to reduce by construction to the fitted parameters or inputs. The framework's physical constraints and MoE routing are presented as modeling choices whose effectiveness is evaluated externally via held-out R², without the derivation chain collapsing into self-definition or fitted-input renaming. This is the common case of an ML paper whose central claims remain falsifiable against independent data splits.

Axiom & Free-Parameter Ledger

3 free parameters · 2 axioms · 1 invented entities

Based solely on the abstract, the framework relies on several fitted components and domain assumptions whose details are not provided; the CPSI appears to be a constructed metric without external validation shown.

free parameters (3)

Expert routing parameters
Learned weights that determine which regime-specific expert activates for given operating conditions
Conservation regularization coefficients
Hyperparameters balancing physical constraint enforcement against data fit
CPSI weighting factors
Coefficients combining carbon and pollutant terms into the synergistic index

axioms (2)

domain assumption Emission processes obey conservation laws that can be expressed as differentiable regularization terms
Invoked to justify physics-informed component of the loss function
domain assumption Plant operating conditions cluster into a finite number of reusable regimes
Basis for the mixture-of-experts architecture and transfer mechanism

invented entities (1)

Carbon-Pollutant Synergistic Index (CPSI) no independent evidence
purpose: Single scalar for integrated risk evaluation of carbon and multiple pollutants
New metric introduced for system-level assessment; no independent external validation or falsifiable prediction provided in abstract

pith-pipeline@v0.9.0 · 5629 in / 1713 out tokens · 76387 ms · 2026-05-07T11:27:31.901803+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

44 extracted references

[1]

Journal of Cleaner Production377, 134501 (2022)

He, R., Sandoval-Reyes, M., Scott, I., Semeano, R., Ferrao, P., Matthews, S., Small, M.J.: Global knowledge base for municipal solid waste management: Framework development and application in waste generation prediction. Journal of Cleaner Production377, 134501 (2022)

2022
[2]

Resources, Conservation and Recycling186, 106528 (2022)

Zhang, C., Dong, H., Geng, Y., Song, X., Zhang, T., Zhuang, M.: Carbon neu- trality prediction of municipal solid waste treatment sector under the shared socioeconomic pathways. Resources, Conservation and Recycling186, 106528 (2022)

2022
[3]

Nature Energy10(2), 215–225 (2025)

Liu, B., Wang, P., Zhou, J., Guo, Y., Ma, S., Chen, W.-Q., Li, J., Chang, V.W.- C.: Refocusing on effectiveness over expansion in urban waste–energy–carbon development in china. Nature Energy10(2), 215–225 (2025)

2025
[4]

Environmental Pollution319, 121021 (2023)

Tang, L., Guo, J., Wan, R., Jia, M., Qu, J., Li, L., Bo, X.: Air pollutant emis- sions and reduction potentials from municipal solid waste incineration in china. Environmental Pollution319, 121021 (2023)

2023
[5]

Environmental Science & Technology58(26), 11342–11351 (2024)

Ma, S., Deng, N., Zhao, C., Wang, P., Zhou, C., Sun, C., Guan, D., Wang, Z., Meng, J.: Decreasing greenhouse gas emissions from the municipal solid waste sector in chinese cites. Environmental Science & Technology58(26), 11342–11351 (2024)

2024
[6]

Journal of Environmental Management 390, 126219 (2025)

Fei, F., Zhang, X., Cheng, Q., Wen, Z., Ma, H., Zhang, J.: New data-driven off-site-supervision mode for flue gas emissions: Data mining of 713 municipal solid-waste incineration factories in china. Journal of Environmental Management 390, 126219 (2025)

2025
[7]

Applied energy 339, 120982 (2023)

Ding, H., Tang, J., Qiao, J.: Dynamic modeling of multi-input and multi-output controlled object for municipal solid waste incineration process. Applied energy 339, 120982 (2023)

2023
[8]

Expert Systems with Applications238, 121783 (2024)

Cui, Y., Meng, X., Qiao, J.: Multi-condition operational optimization with adap- tive knowledge transfer for municipal solid waste incineration process. Expert Systems with Applications238, 121783 (2024)

2024
[9]

Expert Systems with Applications243, 122733 (2024)

Yan, A., Wang, R., Guo, J., Tang, J.: A knowledge transfer online stochastic con- figuration network-based prediction model for furnace temperature in a municipal solid waste incineration process. Expert Systems with Applications243, 122733 (2024)

2024
[10]

Journal of Environmental Management373, 123588 (2025) 20

Qi, Y.-P., He, P.-J., Lan, D.-Y., L¨ u, F., Zhang, H.: Novel method for predicting concentrations of incineration flue gas based on waste composition and machine learning. Journal of Environmental Management373, 123588 (2025) 20

2025
[11]

Energy349, 140644 (2026)

Xu, F., Yin, L., Yang, M., Yin, X., Teng, L., Ye, N., Huang, J., Feng, Y.: A uni- versal hybrid modeling framework for mswi temperature field prediction: fusion of physics-informed and data-driven transfer learning. Energy349, 140644 (2026)

2026
[12]

Engineering Applications of Artificial Intelligence127, 107315 (2024)

Wang, R., Li, F., Yan, A.: Modular stochastic configuration network based pre- diction model for nox emissions in municipal solid waste incineration process. Engineering Applications of Artificial Intelligence127, 107315 (2024)

2024
[13]

Journal of Cleaner Production445, 141313 (2024)

Zhang, R., Tang, J., Xia, H., Chen, J., Yu, W., Qiao, J.: Heterogeneous ensemble prediction model of co emission concentration in municipal solid waste incinera- tion process using virtual data and real data hybrid-driven. Journal of Cleaner Production445, 141313 (2024)

2024
[14]

ACS ES&T Engineering4(3), 737–747 (2024)

Ma, Y., He, P.-J., Lu, F., Zhang, H., Yan, S., Cao, D., Mao, H., Jiang, D.: Machine learning-based prediction of the co2 concentration in the flue gas and carbon emissions from a waste incineration plant. ACS ES&T Engineering4(3), 737–747 (2024)

2024
[15]

In: 2024 6th International Conference on Industrial Artificial Intelligence (IAI), pp

Wang, Z., Tang, J., Xia, H., Zhang, R., Wang, T.: Co2 emission concentration modeling method based on lstm in municipal solid waste incineration process. In: 2024 6th International Conference on Industrial Artificial Intelligence (IAI), pp. 1–4 (2024). IEEE

2024
[16]

Waste Management170, 93–102 (2023)

Wen, C., Lin, X., Ying, Y., Ma, Y., Yu, H., Li, X., Yan, J.: Dioxin emission prediction from a full-scale municipal solid waste incinerator: Deep learning model in time-series input. Waste Management170, 93–102 (2023)

2023
[17]

IEEE Transactions on Industrial Informatics20(1), 358–368 (2023)

Xia, H., Tang, J., Yu, W., Qiao, J.: Online measurement of dioxin emission in solid waste incineration using fuzzy broad learning. IEEE Transactions on Industrial Informatics20(1), 358–368 (2023)

2023
[18]

Journal of Hazardous Materials480, 135834 (2024)

Xia, H., Tang, J., Aljerf, L., Wang, T., Gao, B., Alajlani, M.: Ai-based tree mod- eling for multi-point dioxin concentrations in municipal solid waste incineration. Journal of Hazardous Materials480, 135834 (2024)

2024
[19]

Energy264, 126174 (2023)

Ding, X., Feng, C., Yu, P., Li, K., Chen, X.: Gradient boosting decision tree in the prediction of nox emission of waste incineration. Energy264, 126174 (2023)

2023
[20]

Energy312, 133646 (2024)

Ma, T., Zhou, H., Xu, F., Chen, D., Qian, K., Yin, L.: Numerical simulation and intelligent prediction of a 500 t/d municipal solid waste incinerator. Energy312, 133646 (2024)

2024
[21]

Energy278, 127881 (2023) 21

Yang, Y., Shahbeik, H., Shafizadeh, A., Rafiee, S., Hafezi, A., Du, X., Pan, J., Tabatabaei, M., Aghbashlo, M.: Predicting municipal solid waste gasification using machine learning: A step toward sustainable regional planning. Energy278, 127881 (2023) 21

2023
[22]

Energy243, 122783 (2022)

Tian, L., Lin, K., Zhao, Y., Zhao, C., Huang, Q., Zhou, T.: Combustion perfor- mance of fine screenings from municipal solid waste: Thermo-kinetic investigation and deep learning modeling via tg-ftir. Energy243, 122783 (2022)

2022
[23]

Case Studies in Thermal Engineering61, 105071 (2024)

Lee, D.-S., Lee, S.-T., Chen, Y.-T., Su, P.-Y.: Artificial intelligence technique development for energy-efficient waste-to-energy: A case study of an incineration plant. Case Studies in Thermal Engineering61, 105071 (2024)

2024
[24]

International journal of electrical power & energy systems115, 105508 (2020)

Kabugo, J.C., J¨ ams¨ a-Jounela, S.-L., Schiemann, R., Binder, C.: Industry 4.0 based process data analytics platform: A waste-to-energy plant case study. International journal of electrical power & energy systems115, 105508 (2020)

2020
[25]

Journal of Environmental Management379, 124874 (2025)

Lin, X., Wang, R., Wen, C., Chen, J., Huang, Q., Li, X., Yan, J.: Collaborative prediction and intelligent control of multiple pollutants emission from a large-scale waste incinerator. Journal of Environmental Management379, 124874 (2025)

2025
[26]

Waste Management210, 115256 (2026)

Feng, X., Liu, L., Ye, M., Maˇ sek, O., Gouda, S., Chang, K., Wang, X., Huang, Q.: Unveiling and interpreting the relationships among multi-pollutant emis- sion factors in municipal solid waste incineration by machine learning. Waste Management210, 115256 (2026)

2026
[27]

IEEE Transactions on Industrial Informatics19(12), 11444– 11454 (2023)

Sun, J., Meng, X., Qiao, J.: Data-driven optimal control for municipal solid waste incineration process. IEEE Transactions on Industrial Informatics19(12), 11444– 11454 (2023)

2023
[28]

IEEE Transactions on Industrial Electronics (2025)

Huang, W., Meng, X., Qiao, J.: Cooperative dynamic multiobjective optimiza- tion with multi-time-scale for mswi process. IEEE Transactions on Industrial Electronics (2025)

2025
[29]

Proceedings of the IEEE109(1), 43– 76 (2020)

Zhuang, F., Qi, Z., Duan, K., Xi, D., Zhu, Y., Zhu, H., Xiong, H., He, Q.: A comprehensive survey on transfer learning. Proceedings of the IEEE109(1), 43– 76 (2020)

2020
[30]

IEEE access11, 12887–12910 (2023)

Azari, M.S., Flammini, F., Santini, S., Caporuscio, M.: A systematic literature review on transfer learning for predictive maintenance in industry 4.0. IEEE access11, 12887–12910 (2023)

2023
[31]

Fuel364, 131026 (2024)

Zhu, B., Ren, S., Weng, Q., Si, F.: A physics-informed neural network that consid- ers monotonic relationships for predicting nox emissions from coal-fired boilers. Fuel364, 131026 (2024)

2024
[32]

Modeling Earth Systems and Environment11(4), 297 (2025)

Faiz, Z., Habib, S., Yun, W.: Optimizing plastic waste management and pollu- tion control using a deep neural network-based predictive model. Modeling Earth Systems and Environment11(4), 297 (2025)

2025
[33]

Engineering Applications of 22 Artificial Intelligence164, 113285 (2026)

Adi, V.S.K., Shi, L., Wu, W.: Municipal solid waste gasification predictions using hybrid-data physics-informed neural networks. Engineering Applications of 22 Artificial Intelligence164, 113285 (2026)

2026
[34]

Computer Methods in Applied Mechanics and Engineering429, 117163 (2024)

Taassob, A., Kumar, A., Gitushi, K.M., Ranade, R., Echekki, T.: A pinn-deeponet framework for extracting turbulent combustion closure from multiscalar measure- ments. Computer Methods in Applied Mechanics and Engineering429, 117163 (2024)

2024
[35]

Combustion and Flame269, 113647 (2024)

Zhang, S., Zhang, C., Wang, B.: Crk-pinn: A physics-informed neural network for solving combustion reaction kinetics ordinary differential equations. Combustion and Flame269, 113647 (2024)

2024
[36]

IEEE Industrial Electronics Magazine15(2), 65–75 (2021)

Maschler, B., Weyrich, M.: Deep transfer learning for industrial automation: A review and discussion of new techniques for data-driven machine learning. IEEE Industrial Electronics Magazine15(2), 65–75 (2021)

2021
[37]

IEEE Access12, 173351–173379 (2024)

Yadav, P., Rishiwal, V., Yadav, M., Alotaibi, A., Maurya, V., Agarwal, U., Sharma, S.: Investigation and empirical analysis of transfer learning for industrial iot networks. IEEE Access12, 173351–173379 (2024)

2024
[38]

IEEE Transactions on Automation Science and Engineering22, 9338–9352 (2024)

Qiao, J., Cui, Y., Meng, X.: Dynamic multi-objective operation optimization of municipal solid waste incineration process based on transfer learning. IEEE Transactions on Automation Science and Engineering22, 9338–9352 (2024)

2024
[39]

Green Energy and Resources2(3), 100086 (2024)

Chen, G., Chen, G., Li, J., Pan, Q., Liang, D., Qiu, J., Zhao, X., Wang, X., Li, Z., Li, X.,et al.: Co-incineration of multiple inorganic solid wastes towards clean disposal: Heat and mass transfer modeling, pollutant generation, and machine learning based proportioning. Green Energy and Resources2(3), 100086 (2024)

2024
[40]

Nature Communications15(1), 10700 (2024)

Koldasbayeva, D., Tregubova, P., Gasanov, M., Zaytsev, A., Petrovskaia, A., Burnaev, E.: Challenges in data-driven geospatial modeling for environmental research and practice. Nature Communications15(1), 10700 (2024)

2024
[41]

In: International Conference on Machine Learning, pp

Rajbhandari, S., Li, C., Yao, Z., Zhang, M., Aminabadi, R.Y., Awan, A.A., Rasley, J., He, Y.: Deepspeed-moe: Advancing mixture-of-experts inference and training to power next-generation ai scale. In: International Conference on Machine Learning, pp. 18332–18346 (2022). PMLR

2022
[42]

Advances in neural information processing systems34, 15908–15919 (2021)

Han, K., Xiao, A., Wu, E., Guo, J., Xu, C., Wang, Y.: Transformer in transformer. Advances in neural information processing systems34, 15908–15919 (2021)

2021
[43]

Neural computation31(7), 1235–1270 (2019)

Yu, Y., Si, X., Hu, C., Zhang, J.: A review of recurrent neural networks: Lstm cells and network architectures. Neural computation31(7), 1235–1270 (2019)

2019
[44]

International Journal of Robotics and Control Systems2(4), 739–748 (2022) 23

Purwono, P., Ma’arif, A., Rahmaniar, W., Fathurrahman, H.I.K., Frisky, A.Z.K., Haq, Q.M.: Understanding of convolutional neural network (cnn): A review. International Journal of Robotics and Control Systems2(4), 739–748 (2022) 23

2022