pith. sign in

arxiv: 2605.04068 · v1 · submitted 2026-04-12 · 💻 cs.LG · cs.AI

Designing a double deep reinforcement learning selection tool for resilient demand prediction

Bilel Abderrahmane Benziane , Benoit Lardeux , Ayoub Mcharek , Maher Jridi This is my paper

Pith reviewed 2026-05-10 15:43 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords deep reinforcement learningdemand forecastingmodel selectionsupply chainearly stoppingtime series prediction
0
0 comments X

The pith

A double deep reinforcement learning agent automatically selects forecasting models from a committee at prediction time for demand data.

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

The paper proposes a double deep reinforcement learning architecture that learns to pick the best model from a forecasting committee dynamically when making predictions rather than fixing the choice in advance. It adds an early-stopping rule based on average reward convergence to shorten training. Tests on grocery sales and snack demand datasets indicate the approach holds up better or as well as existing methods. A reader would care because choosing the right forecasting model by hand is hard when datasets differ in their patterns, and automation could make supply chain planning more reliable. If the claim holds, reinforcement learning becomes a practical way to handle model selection without constant expert tuning.

Core claim

The central claim is that a double DRL agent can learn policies to choose forecasting models from a committee at the moment of prediction, paired with reward-based early stopping, and that this yields robust results on real grocery and snack demand datasets compared with prior selection techniques.

What carries the argument

The double deep reinforcement learning agent that decides which model from the forecasting committee to apply, guided by policies trained on prediction rewards.

If this is right

  • Selection occurs at prediction time instead of requiring a fixed choice before any data arrives.
  • The average-reward early-stopping shortens the training phase while preserving final performance.
  • The same selector architecture applies across grocery sales and snack demand data with consistent robustness gains.
  • No manual intervention is needed to switch models when dataset characteristics change.

Where Pith is reading between the lines

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

  • The selector could be tested on other time-series tasks such as energy load or inventory forecasting to check broader usefulness.
  • Integrating the DRL selector with larger model committees might further improve resilience when data patterns shift suddenly.
  • The approach might lower the expertise barrier for companies that lack dedicated forecasting teams.

Load-bearing premise

The double DRL agent learns selection policies that work on new or unseen demand datasets without overfitting to the ones used for training.

What would settle it

Running the method on a fresh demand dataset from a different domain or time period and checking whether its accuracy or robustness falls below that of standard model-selection baselines.

Figures

Figures reproduced from arXiv: 2605.04068 by Ayoub Mcharek, Benoit Lardeux, Bilel Abderrahmane Benziane, Maher Jridi.

Figure 1
Figure 1. Figure 1: Double deep Q learning architecture. In Double Deep Q Learning (DDQN), the process involves several key elements: 1. Q-Networks: (a) Online Q-Network: This neural network estimates Q-values, which represent the expected cumulative rewards for taking various actions in the current state. It’s updated during the training process. 7 [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The suggested CRFFNN architecture. smoothing equation is given by equation (3): srt = 1 n nX−1 i=0 rt−i (3) where srt represents the smoothed reward point at time step t, n is the window size, and rt are the original reward points within the window centered around time step t. The second part involves the implementation of the actual early stopping mechanism, which incorporates a patience counter. This cou… view at source ↗
Figure 3
Figure 3. Figure 3: Average reward improvement rate based on early forecasting plot. [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The experimental setup. selected, this model is trained again on both the train and validation sets. The forecasting error is finally measured on the test data set. In both datasets, to fill in missing values, a straightforward substitution method was employed, replacing all missing values with zeros. In contrast to previous research highlighting the importance of unseasonalization [41], this study refrain… view at source ↗
Figure 5
Figure 5. Figure 5: MASE and SMAPE Errors for NN5 Dataset To validate the usefulness of the suggested early stopping methods ARIBES, The same suggested CRFFNN method is implemented with early stopping CRFFNN-ARIBES. The training time and forecast errors are summarized in [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: MASE and SMAPE Errors for Case Study Dataset [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗
read the original abstract

The use of artificial intelligence in supply chain forecasting has attracted many scientific studies for several decades. However, the process of selecting an appropriate forecasting solution becomes a daunting task. This complexity arises due to the distinct features inherent to each dataset. Research to tackle this issue has been performed since the eighties but recent development of demand forecasting has opened new perspectives. This research aims to enhance automatic forecasting model selection by proposing a novel architecture that acts as a double deep reinforcement learning agent, selecting automatically a forecasting model from the forecasting committee at the time of prediction. Moreover, a novel early-stopping approach based on average reward convergence has been introduced to expedite training time. To evaluate the model's performance, an empirical study was conducted utilizing grocery sales datasets and snack demands datasets. The experimental results demonstrate the robustness of the proposed approach when compared to state-of-the-art methods.

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.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes a double deep reinforcement learning (DRL) agent that automatically selects a forecasting model from a committee at prediction time for demand forecasting in supply chains. It introduces a novel early-stopping criterion based on average reward convergence to accelerate training. The approach is evaluated empirically on grocery sales and snack demand datasets, with the central claim being that it demonstrates robustness relative to state-of-the-art methods.

Significance. If the robustness and generalization claims are substantiated, the work could offer a practical advance in automated, adaptive model selection for time-series demand forecasting, potentially improving resilience in supply-chain applications. The double-DRL selector and reward-convergence early stopping are conceptually interesting contributions that address a real operational pain point.

major comments (2)
  1. [Experimental results] Experimental results section: the abstract and evaluation description assert robustness on grocery sales and snack demand datasets versus SOTA, yet provide no details on experimental setup, chosen baselines, metrics, statistical significance testing, train/test splits, or any cross-dataset hold-out protocol. This leaves the central generalization claim unsupported and prevents assessment of whether the policy learns transferable selection rules or merely memorizes dataset-specific patterns.
  2. [Methodology] Methodology section: the double DRL architecture is presented at a high level without explicit definitions of state space, action space, reward function, or the interaction between the two agents. Absent these (or pseudocode/equations), it is impossible to verify the claimed novelty or to reproduce the selection policy.
minor comments (1)
  1. [Abstract] Abstract: the phrase 'robustness of the proposed approach' should be qualified with the specific metrics (e.g., MAE, RMSE, or selection accuracy) on which superiority is claimed.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful and constructive report. We address each major comment below and will revise the manuscript to provide the requested details, which we agree will improve clarity, reproducibility, and support for the central claims.

read point-by-point responses

  1. Referee: [Experimental results] Experimental results section: the abstract and evaluation description assert robustness on grocery sales and snack demand datasets versus SOTA, yet provide no details on experimental setup, chosen baselines, metrics, statistical significance testing, train/test splits, or any cross-dataset hold-out protocol. This leaves the central generalization claim unsupported and prevents assessment of whether the policy learns transferable selection rules or merely memorizes dataset-specific patterns.

    Authors: We agree that additional experimental details are required to fully substantiate the robustness and generalization claims. In the revised manuscript we will expand the Experimental Results section with: (i) a complete description of the experimental setup and data preprocessing; (ii) the full list of baselines (individual forecasters plus competing selection methods); (iii) the precise metrics (MAE, RMSE, MAPE) and any secondary measures; (iv) statistical significance testing (paired t-tests or Wilcoxon signed-rank tests with p-values); (v) explicit train/test split ratios, temporal ordering, and any cross-dataset hold-out protocol. These additions will allow readers to evaluate whether the double-DRL policy learns transferable selection rules across the grocery-sales and snack-demand domains. revision: yes

  2. Referee: [Methodology] Methodology section: the double DRL architecture is presented at a high level without explicit definitions of state space, action space, reward function, or the interaction between the two agents. Absent these (or pseudocode/equations), it is impossible to verify the claimed novelty or to reproduce the selection policy.

    Authors: We acknowledge that the current description of the double-DRL architecture is high-level. In the revised version we will add: (i) formal definitions of the state space (historical demand statistics, dataset meta-features, and prediction-time context), action space (discrete selection from the forecasting committee), and reward function (negative forecast error plus a small regularization term); (ii) the precise interaction protocol between the two agents (one performing model selection, the second providing auxiliary policy guidance); and (iii) pseudocode together with the key Q-learning or policy-gradient update equations. These additions will make the novelty of the average-reward-convergence early-stopping criterion and the overall architecture fully verifiable and reproducible. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical proposal with no derivations or self-referential reductions

full rationale

The paper introduces a double deep reinforcement learning agent for automatic forecasting model selection and reports empirical results on grocery sales and snack demand datasets. No equations, derivations, or parameter-fitting steps appear in the abstract or described content. The central claim of robustness versus SOTA rests on experimental comparisons rather than any self-definitional loop, fitted input renamed as prediction, or load-bearing self-citation chain. The evaluation uses internal splits and early stopping but does not reduce any asserted generalization to a tautology by construction; therefore the derivation chain (such as it is) is self-contained and non-circular.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No free parameters, axioms, or invented entities are described in the abstract; the approach appears to rely on standard RL components and existing forecasting models without new postulates.

pith-pipeline@v0.9.0 · 5451 in / 988 out tokens · 24422 ms · 2026-05-10T15:43:43.253138+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

51 extracted references · 51 canonical work pages

  1. [1]

    Impact of covid-19 on demand planning: Building resilient forecasting models

    Sreeja Ashok and Kanu Aravind. Impact of covid-19 on demand planning: Building resilient forecasting models. InProceedings of the 2021 5th International Conference on Compute and Data Analysis, ICCDA ’21, page 59–66, New York, NY, USA, 2021. Association for Computing Machinery

    work page 2021

  2. [2]

    Forecasting is difficult, especially about the future: Using contentious issues to forecast interstate disputes.Journal of Peace Research, 50(1):17–31, 2013

    Kristian Skrede Gleditsch and Michael D Ward. Forecasting is difficult, especially about the future: Using contentious issues to forecast interstate disputes.Journal of Peace Research, 50(1):17–31, 2013

    work page 2013

  3. [3]

    Demand forecasting using random forest and artificial neural network for supply chain management

    Navneet Vairagade, Doina Logofatu, Florin Leon, and Fitore Muharemi. Demand forecasting using random forest and artificial neural network for supply chain management. In Ngoc Thanh Nguyen, 20 Richard Chbeir, Ernesto Exposito, Philippe Aniorté, and Bogdan Trawiński, editors,Computational Collective Intelligence, pages 328–339, Cham, 2019. Springer Internat...

    work page 2019

  4. [4]

    Artificial Intelligence in Supply Chain Management: Investigation of Transfer Learning to Improve Demand Forecasting of Intermittent Time Series with Deep Learning

    Daniel Kiefer, Florian Grimm, and Dinther Van. Artificial Intelligence in Supply Chain Management: Investigation of Transfer Learning to Improve Demand Forecasting of Intermittent Time Series with Deep Learning. InProceedings of the 55th Hawaii International Conference on System Sciences, 2022

    work page 2022

  5. [5]

    Koushiki Dasgupta Chaudhuri and Bugra Alkan. A hybrid extreme learning machine model with harris hawks optimisation algorithm: an optimised model for product demand forecasting applications.Applied Intelligence, 52(10):11489–11505, 2022

    work page 2022

  6. [6]

    Abhishekh, Surendra Singh Gautam, and S. R. Singh. A new method of time series forecasting using intuitionistic fuzzy set based on average-length.Journal of Industrial and Production Engineering, May

    work page

  7. [7]

    Publisher: Taylor & Francis

    work page

  8. [8]

    Supply Chain Forecasting in a fast-moving global economy: Review, Limits and Future Directions

    Bilel Abderrahmane Benziane, Benoit Lardeux, Maher Jridi, and Ayoub Mcharek. Supply Chain Forecasting in a fast-moving global economy: Review, Limits and Future Directions. InThe 28th International Conference on Automation and Computing, Birmingham, United Kingdom, August 2023. IEEE Robotics & Automation Society

    work page 2023

  9. [9]

    Investigating explanatory variables impact on warehouse demand forecasting

    Bilel Abderrahmane Benziane, Benoit Lardeux, Maher Jridi, and Ayoub Mcharek. Investigating explanatory variables impact on warehouse demand forecasting. In2024 IEEE/ACS 21st International Conference on Computer Systems and Applications (AICCSA), pages 1–6, 2024

    work page 2024

  10. [10]

    Does a higher number of parameters necessarily mean lower forecasting error?TechRxiv, 2025(0605), 2025

    Bilel Abderrahmane BENZIANE, Benoit Lardeux, Maher Jridi, Ayoub Mcharek, and Xavier Schepler. Does a higher number of parameters necessarily mean lower forecasting error?TechRxiv, 2025(0605), 2025

    work page 2025

  11. [11]

    An ensemble framework for probabilistic short-term load forecasting based on bitcn and deep attention networks.TechRxiv, 2025(0319), 2025

    Bilel Abderrahmane Benziane, Benoit Lardeux, Maher Jridi, and Ayoub Mcharek. An ensemble framework for probabilistic short-term load forecasting based on bitcn and deep attention networks.TechRxiv, 2025(0319), 2025

    work page 2025

  12. [12]

    Theses, Université de Brest, June 2025

    Bilel Abderrahmane Benziane.Artificial inteligence in supply chain demand forecasting. Theses, Université de Brest, June 2025

    work page 2025

  13. [13]

    Dealing with uncertainty in the supply chain through intelligent optimization methods

    Khadija Hadj Salem, Benziane Bilel Abderrahmane, Hammami Nour El Houda, Lardeux Benoit, Schepler Xavier, Hadj-Alouane Atidel B., and Jridi Maher. Dealing with uncertainty in the supply chain through intelligent optimization methods. InForging Bridges between Artificial Intelligence and Operations Research: Applications in Healthcare and Supply Chain Manag...

    work page 2025

  14. [14]

    Wenhan Fu and Chen-Fu Chien. Unison data-driven intermittent demand forecast framework to empower supply chain resilience and an empirical study in electronics distribution.Computers and Industrial Engineering, 135:940–949, 2019

    work page 2019

  15. [15]

    Applying the zero-inflated Poisson regression in the inventory management of irregular demand items.Journal of Industrial and Production Engineering, 39(6):458–478, August 2022

    Serena Finco, Daria Battini, Giuseppe Converso, and Teresa Murino. Applying the zero-inflated Poisson regression in the inventory management of irregular demand items.Journal of Industrial and Production Engineering, 39(6):458–478, August 2022. Publisher: Taylor & Francis _eprint: https://doi.org/10.1080/21681015.2022.2041741

    work page doi:10.1080/21681015.2022.2041741 2022

  16. [16]

    Demand forecasting in supply chain: The impact of demand volatility in the presence of promotion.Computers and Industrial Engineering, 142:106380, 2020

    Mahdi Abolghasemi, Eric Beh, Garth Tarr, and Richard Gerlach. Demand forecasting in supply chain: The impact of demand volatility in the presence of promotion.Computers and Industrial Engineering, 142:106380, 2020

    work page 2020

  17. [17]

    Murray, Bruno Agard, and Marco A

    Paul W. Murray, Bruno Agard, and Marco A. Barajas. Forecast of individual customer’s demand from a large and noisy dataset.Computers and Industrial Engineering, 118:33–43, 2018

    work page 2018

  18. [18]

    Optimisation of water demand forecasting by artificial intelligence with short data sets.Biosystems Engineering, 177:59–66, 2019

    Rafael González Perea, Emilio Camacho Poyato, Pilar Montesinos, and Juan Antonio RodrÃguez DÃaz. Optimisation of water demand forecasting by artificial intelligence with short data sets.Biosystems Engineering, 177:59–66, 2019. Intelligent Systems for Environmental Applications. 21

    work page 2019

  19. [19]

    Review and analysis of artificial intelligence methods for demand forecasting in supply chain management.Procedia CIRP, 107, 2022

    Mario Angos Mediavilla, Fabian Dietrich, and Daniel Palm. Review and analysis of artificial intelligence methods for demand forecasting in supply chain management.Procedia CIRP, 107, 2022

    work page 2022

  20. [20]

    Warehouse demand forecasting based on long short-term memory neural networks

    Kerim Hodžić, Haris Hasić, Emir Cogo, and Željko Jurić. Warehouse demand forecasting based on long short-term memory neural networks. In2019 XXVII International Conference on Information, Communication and Automation Technologies (ICAT), pages 1–6, 2019

    work page 2019

  21. [21]

    Electricity load forecasting for each day of week using deep cnn

    Sajjad Khan, Nadeem Javaid, Annas Chand, Abdul Basit Majeed Khan, Fahad Rashid, and Imran Uddin Afridi. Electricity load forecasting for each day of week using deep cnn. In Leonard Barolli, Makoto Tak- izawa, Fatos Xhafa, and Tomoya Enokido, editors,Web, Artificial Intelligence and Network Applications, pages 1107–1119, Cham, 2019. Springer International ...

    work page 2019

  22. [22]

    A short-term load forecasting method using integrated cnn and lstm network.IEEE Access, 9:32436–32448, 2021

    Shafiul Hasan Rafi, Nahid-Al-Masood, Shohana Rahman Deeba, and Eklas Hossain. A short-term load forecasting method using integrated cnn and lstm network.IEEE Access, 9:32436–32448, 2021

    work page 2021

  23. [23]

    Sainath, Oriol Vinyals, Andrew Senior, and Ha?im Sak

    Tara N. Sainath, Oriol Vinyals, Andrew Senior, and Ha?im Sak. Convolutional, long short-term memory, fully connected deep neural networks. In2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 4580–4584, 2015

    work page 2015

  24. [24]

    Multi-Step Short-Term Power Consumption Forecasting with a Hybrid Deep Learning Strategy.Energies, 11(11):3089, November

    Ke Yan, Xudong Wang, Yang Du, Ning Jin, Haichao Huang, and Hangxia Zhou. Multi-Step Short-Term Power Consumption Forecasting with a Hybrid Deep Learning Strategy.Energies, 11(11):3089, November

    work page

  25. [25]

    Number: 11 Publisher: Multidisciplinary Digital Publishing Institute

    work page

  26. [26]

    A hybrid deep learning framework with cnn and bi-directional lstm for store item demand forecasting.Computers and Electrical Engineering, 103:108358, 2022

    Reuben Varghese Joseph, Anshuman Mohanty, Soumyae Tyagi, Shruti Mishra, Sandeep Kumar Satapathy, and Sachi Nandan Mohanty. A hybrid deep learning framework with cnn and bi-directional lstm for store item demand forecasting.Computers and Electrical Engineering, 103:108358, 2022

    work page 2022

  27. [27]

    R. G. Brown. Exponential smoothing for predicting demand. 1956

    work page 1956

  28. [28]

    George E. P. Box and Gwilym M. Jenkins.Time Series Analysis: Forecasting and Control. Holden-Day, 1970

    work page 1970

  29. [29]

    Crone, Michèle Hibon, and Konstantinos Nikolopoulos

    Sven F. Crone, Michèle Hibon, and Konstantinos Nikolopoulos. Advances in forecasting with neural networks? Empirical evidence from the NN3 competition on time series prediction.International Journal of Forecasting, 27(3), 2011

    work page 2011

  30. [30]

    ?lker Güven and Fuat Şimşir. Demand forecasting with color parameter in retail apparel industry using artificial neural networks (ANN) and support vector machines (SVM) methods.Computers & Industrial Engineering, 147, 2020

    work page 2020

  31. [31]

    Long short-term memory.Neural computation, 9:1735–80, 1997

    Sepp Hochreiter and J�rgen Schmidhuber. Long short-term memory.Neural computation, 9:1735–80, 1997

    work page 1997

  32. [32]

    Erdin�Ko�and Muammer T�rko?lu. Forecasting of medical equipment demand and outbreak spreading based on deep long short-term memory network: the COVID-19 pandemic in Turkey.Signal, Image and Video Processing, 16(3):613–621, 2022

    work page 2022

  33. [33]

    Naraganahalli

    Kiran Kumar Chandriah and Raghavendra V. Naraganahalli. RNN / LSTM with modified Adam optimizer in deep learning approach for automobile spare parts demand forecasting.Multimedia Tools and Applications, 80(17):26145–26159, 2021

    work page 2021

  34. [34]

    Forecasting across time series databases using recurrent neural networks on groups of similar series: A clustering approach.Expert Systems with Applications, 140:112896, 2020

    Kasun Bandara, Christoph Bergmeir, and Slawek Smyl. Forecasting across time series databases using recurrent neural networks on groups of similar series: A clustering approach.Expert Systems with Applications, 140:112896, 2020

    work page 2020

  35. [35]

    Combination of short-term load forecasting models based on a stacking ensemble approach.Energy and Buildings, 216:109921, 2020

    Jihoon Moon, Seungwon Jung, Jehyeok Rew, Seungmin Rho, and Eenjun Hwang. Combination of short-term load forecasting models based on a stacking ensemble approach.Energy and Buildings, 216:109921, 2020. 22

    work page 2020

  36. [36]

    An optimized model using lstm network for demand forecasting.Computers and Industrial Engineering, 143:106435, 2020

    Hossein Abbasimehr, Mostafa Shabani, and Mohsen Yousefi. An optimized model using lstm network for demand forecasting.Computers and Industrial Engineering, 143:106435, 2020

    work page 2020

  37. [37]

    Villegas, Diego J

    Marco A. Villegas, Diego J. Pedregal, and Juan R. Trapero. A support vector machine for model selection in demand forecasting applications.Computers and Industrial Engineering, 121:1–7, 2018

    work page 2018

  38. [38]

    Reinforcement learning- based load forecasting of electric vehicle charging station using q-learning technique.IEEE Transactions on Industrial Informatics, 17(6):4229–4237, 2021

    Morteza Dabbaghjamanesh, Amirhossein Moeini, and Abdollah Kavousi-Fard. Reinforcement learning- based load forecasting of electric vehicle charging station using q-learning technique.IEEE Transactions on Industrial Informatics, 17(6):4229–4237, 2021

    work page 2021

  39. [39]

    Mahmassani, and Ying Chen

    Lama Al Hajj Hassan, Hani S. Mahmassani, and Ying Chen. Reinforcement learning framework for freight demand forecasting to support operational planning decisions.Transportation Research Part E: Logistics and Transportation Review, 137:101926, 2020

    work page 2020

  40. [40]

    Human-level control through deep reinforcement learning.Nature, 518:529–33, 02 2015

    Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei Rusu, Joel Veness, Marc Bellemare, Alex Graves, Martin Riedmiller, Andreas Fidjeland, Georg Ostrovski, Stig Petersen, Charles Beattie, Amir Sadik, Ioannis Antonoglou, Helen King, Dharshan Kumaran, Daan Wierstra, Shane Legg, and Demis Hassabis. Human-level control through deep reinforcement learning.N...

    work page 2015

  41. [41]

    Deep reinforcement learning with double q-learning

    Hado Van Hasselt, Arthur Guez, and David Silver. Deep reinforcement learning with double q-learning. Proceedings of the AAAI Conference on Artificial Intelligence, 30, 09 2015

    work page 2015

  42. [42]

    Recurrent neural networks for time series forecasting: Current status and future directions.International Journal of Forecasting, 37(1):388–427, 2021

    Hansika Hewamalage, Christoph Bergmeir, and Kasun Bandara. Recurrent neural networks for time series forecasting: Current status and future directions.International Journal of Forecasting, 37(1):388–427, 2021

    work page 2021

  43. [43]

    Data pre-processing for neural network-based forecasting: does it really matter?Technological and Economic Development of Economy, 23(5):709–725, June 2017

    Oscar Claveria, Enric Monte, and Salvador Torra. Data pre-processing for neural network-based forecasting: does it really matter?Technological and Economic Development of Economy, 23(5):709–725, June 2017. Number: 5

    work page 2017

  44. [44]

    Kaggle Web Traffic Time Series Forecasting, September 2023

    Arthur Suilin. Kaggle Web Traffic Time Series Forecasting, September 2023. original-date: 2017-11- 17T21:15:59Z

    work page 2023

  45. [45]

    Rosenblatt

    F. Rosenblatt. The perceptron: A probabilistic model for information storage and organization in the brain.Psychological Review, 65(6):386–408, 1958. Place: US Publisher: American Psychological Association

    work page 1958

  46. [46]

    Learning phrase representations using RNN encoder–decoder for statistical machine translation

    Kyunghyun Cho, Bart van Merriënboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. Learning phrase representations using RNN encoder–decoder for statistical machine translation. InProceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1724–1734, Doha, Qatar, October 201...

    work page 2014

  47. [47]

    Bidirectional recurrent neural networks.Signal Processing, IEEE Transactions on, 45:2673 – 2681, 12 1997

    Mike Schuster and Kuldip Paliwal. Bidirectional recurrent neural networks.Signal Processing, IEEE Transactions on, 45:2673 – 2681, 12 1997

    work page 1997

  48. [48]

    Gradient-Based Learning Applied to Document Recognition

    Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Ha. Gradient-Based Learning Applied to Document Recognition. 1998

    work page 1998

  49. [49]

    Bishop.Pattern recognition and machine learning

    Christopher M. Bishop.Pattern recognition and machine learning. Information science and statistics. Springer, New York, 2006

    work page 2006

  50. [50]

    Perrone and Leon N

    Michael P. Perrone and Leon N. Cooper. When networks disagree: Ensemble methods for hybrid neural networks. InHow We Learn; How We Remember: Toward an Understanding of Brain and Neural Systems, volume Volume 10 ofWorld Scientific Series in 20th Century Physics, pages 342–358. WORLD SCIENTIFIC, September 1995

    work page 1995

  51. [51]

    David H. Wolpert. Stacked generalization.Neural Networks, 5(2):241–259, January 1992. 23

    work page 1992