Pith. sign in

REVIEW 4 major objections 5 minor 155 references

A Predict-then-Correct Loop Based on Few-Shot Continuous Contextual Bandit for Demand Forecasting

T0 review · 4 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read A lightweight correction layer that adjusts base ML forecasts in real time can cut retail forecast error and inventory cost without retraining the underlying model.

desk verdict Useful framing and a plausible mechanism, but the significance claim is unspecified and the top-p ratio is tuned on the eval window — treat the error reductions as unverified until the authors release code/data. read the letter →

arxiv 2607.16354 v1 pith:VPHVRT6B submitted 2026-07-17 cs.LG cs.AI

Zhiwei Lei , Benedict Jun Ma , Ilya Jackson This is my paper · ORCID
classification cs.LGcs.AI
keywords demandforecastingcontextualbanditfew-shotlearningpredict-then-correcttop-pparameterupdateretailanalyticsinventorymanagement
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

The paper tries to establish that demand forecasts need not be replaced when they drift; a small online correction layer can fix them. It proposes a predict-then-correct loop in which an offline machine-learning model makes the base forecast and a few-shot continuous contextual bandit chooses a multiplicative adjustment to that forecast based on recent context. The reward is simply how much the adjustment shrinks the absolute forecast error, and the policy updates only a small top-p fraction of its parameters to avoid overfitting sparse new-cycle data. Across Walmart and beverage-industry data, the corrected forecasts significantly reduce MAPE, MAE, and RMSE for stable and erratic demand patterns, and the corrected forecasts also produce lower inventory costs than base-stock, PPO, and SAC policies under the tested lead times. If right, this means early-cycle and low-volume items can adapt to changing demand without expensive full retraining.

What carries the argument

The engine is the few-shot continuous contextual bandit (FSCCB) correction layer: a small policy network that outputs a Gaussian correction ratio, trained on a sliding window of (context, action, reward) tuples, updated by a top-p masked rule that updates only the smallest-magnitude parameters, and supported by similar-SKU augmentation. The load-bearing identity is corrected forecast = ML forecast x (1+a) with reward = scaling factor x (|ML error| - |corrected error|), and the containment argument that the zero-correction action is in the policy class.

What would settle it

Run the same loop on demand that is revealed only after a multi-day delay, so that the reward at forecast time cannot use the actual demand. If corrected RMSE and inventory costs are no better than using the base ML forecast alone, the loop's benefit is tied to immediate feedback and the central claim fails.

Watch

Extended reading notes

Core claim

A second-stage correction policy can adapt a frozen base forecast to new demand cycles. It reads recent actuals, base errors, and calendar features, samples an adjustment a in [-1,2], and outputs base forecast x (1+a). Reward is the reduction in absolute error versus the base, so maximizing reward is minimizing corrected error. Since a zero-correction action is in the policy class, optimal population loss never exceeds the base forecast. Empirically, corrections significantly cut MAPE, MAE, and RMSE across stable and erratic demand on Walmart and beverage data, with a 9.52% RMSE gain over the ML-only baseline in the ablation.

Load-bearing premise

The loop assumes true demand is observed immediately after each forecast, so the reward can be computed; when demand feedback is delayed or aggregated, the correction policy cannot learn.

Editorial extensions

If this is right

  • Retailers can deploy the correction layer on top of existing ML forecasters and keep the base model frozen, cutting retraining cost and time.
  • Sparse early-cycle or cold-start SKUs can be corrected using cluster-neighbor augmentation and selective parameter updates instead of waiting for enough labels.
  • Because zero correction belongs to the policy class, the corrected system is guaranteed in the population limit to match, not worsen, the base forecast.
  • Corrected forecasts reduce expected inventory holding and shortage costs relative to base-stock, PPO, and SAC in the tested periodic-review settings.
  • The ablation indicates that full-parameter correction barely helps (0.34% RMSE gap) while top-p selective correction gives 9.52%, so the masking mechanism is the source of the gain.

Reading between the lines

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

  • Beyond the paper: a deployment-ready version would need an automatic way to set the update ratio p; the paper tunes it per dataset (0.10 Walmart, 0.16 beverage), so the reported gains partly reflect that tuning.
  • Beyond the paper: the one-step absolute-error reward is a proxy; extending the reward to expected inventory cost or service level would test whether the corrected forecasts optimize the actual decision objective.
  • Beyond the paper: delayed demand feedback, which the authors list as future work, breaks the immediate-reward loop; a delayed-reward estimator would be needed for real retail settings.
  • Beyond the paper: similar-SKU augmentation assumes cluster neighbors are safe to borrow from; with strong substitution or promotion effects, a graph-aware augmentation would be a testable safer extension.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes a two-stage 'predict-then-correct' (PtC) loop: an offline ML base forecaster is retained, and a continuous contextual bandit policy with top-p masked updates corrects the base forecast using a few-shot sliding-window feedback set. Similar-SKU augmentation is used to expand sparse data. The authors prove three elementary propositions (a containment result, a plasticity-stability tradeoff, and a descent lemma for the masked update) and report empirical gains on Walmart M5 and a proprietary beverage dataset, plus a downstream inventory-cost comparison. The headline claim is statistically significant reductions in MAPE/MAE/RMSE across three demand categories and lower inventory cost than BS/PPO/SAC.

Significance. If the empirical results were valid, the modular correction layer would be a useful contribution because it avoids retraining the base model and directly couples forecasting to downstream decisions. The ablation isolating FSCCB from plain CB is informative, and the theoretical propositions are stated with honest caveats about their scope. However, the paper currently ships no code or data, the significance claims are not backed by any described test, and the main hyperparameter appears to be tuned on the evaluation datasets. These issues are load-bearing, so the central claim is not yet established.

major comments (4)
  1. [§4.2.1, Tables 2–3, §4.3, Appendix B] The sentence 'all observed improvements are statistically significant at the p<0.01 level' is not supported by any described test. No paired test statistic, no statement about whether the test is across SKUs or forecast cycles, and no multiple-comparison correction for the nine metric-by-category comparisons is given. More importantly, Appendix B sets the top-p update ratio per dataset (0.10 for Walmart, 0.16 for beverage) and §4.3 selects p as the peak of the improvement curve on each dataset. If p is chosen on the same forecast window on which Tables 2–3 are computed, the reported reductions are maxima over a p-grid and the p<0.01 stars are not valid. The authors must either fix p on validation data before the evaluation window or provide a bias-corrected evaluation, and must describe the test fully.
  2. [Table 3] Several baseline cells are reported as '–' without explanation (ETS for Stable & Low Volume and Erratic & Intermittent; LightGBM for Stable & High Volume and Erratic & Intermittent). This makes the claim of consistent significant improvement across all conditions unverifiable and the comparison uneven. Full results or an explicit reason for omission should be reported.
  3. [§5.3, Figure 5] The abstract and §5 claim PtC 'yields lower inventory costs' than BS, PPO, and SAC, but the only evidence is a qualitative figure. No numerical costs, standard errors/confidence intervals, or number of replications are reported. Given the stochasticity of the inventory simulation and RL baselines, the claim needs a quantitative table and significance statement or should be withdrawn.
  4. [§3.2.2, Eq. (6), §6] The loop requires observing actual demand y_t immediately after each forecast to compute r_t. The conclusion lists delayed feedback as future work, which is appropriate, but the abstract's 'real-time retail decision-making' overstates the current setting. Please clarify the assumption of immediate feedback in the problem formulation and scope the claims accordingly.
minor comments (5)
  1. [§3.2.4, Eq. (14)] The mask notation uses \theta_k^j before defining the layer index k; define the ordering explicitly. Also, 'smallest p% parameters' is ambiguous when p is a proportion rather than an integer count.
  2. [§3.1 and Appendix B] The similar-SKU augmentation parameters k, distance threshold d, and minimum training size are not listed in the hyperparameter table. Please report them for reproducibility.
  3. [Appendix A] The Walmart forecast window is only 9 weeks (weeks 144–152). The paper should state whether results are stable across different window lengths or justify this choice.
  4. [Data Availability] The statement 'data available upon reasonable request' and the absence of code make the empirical results difficult to verify. Releasing code and anonymized data (at least for the public Walmart/M5 portion) would substantially strengthen the paper.
  5. [Throughout] Minor typographical issues: 'd is is' in §3.1, 'Lipchitz' in Appendix E, 'F ew-Shot' in the supplement header, and inconsistent use of 'top-p' versus 'Top-p'.

Circularity Check

2 steps flagged · score 4.0 of 10

Definitional Proposition 1 and test-set-tuned top-p make some claimed guarantees vacuous and part of the reported gains fitted, but the empirical core is not fully circular.

  1. self definitional [Section 3.2.3, Proposition 1 and Appendix D]
    "Let L∗PtC and LML denote the optimal corrected and ML-only expected losses induced by any nonnegative forecasting loss ℓ. If 0∈A and π0(s)=0∈Π, then L∗PtC≤LML. ... Since 0∈A and π0(st)=0 belongs to the policy class Π, the PtC framework can choose the zero-correction policy... the ML-only prediction is a special case of the PtC prediction class."

    Proposition 1 is true by construction: the PtC policy class is defined to contain the zero-correction policy π0(s)=0, and the PtC loss is the same forecasting loss evaluated on ŷ_ML(1+π(s)) versus y. Taking the infimum over a class that already contains the ML-only policy cannot exceed the loss of that policy. The proof reduces immediately to the definition of the policy class and the loss; it is a tautological containment property, not an independent guarantee. The paper itself acknowledges it is only an 'expressiveness and containment property,' but still states it as a Proposition and a formal result.

  2. fitted input called prediction [Section 4.3, Figure 3; Appendix B, Table B1]
    "On the beverage-company dataset, the improvement rises rapidly from a small update proportion and reaches its peak p=0.16... On the Walmart dataset, the improvement is already significant under small update proportions and reaches its highest value p=0.10... Parameter update ratio for the Walmart dataset 0.10; Parameter update ratio for the beverage dataset 0.16."

    The top-p update ratio is selected per dataset by maximizing the reported improvement on the evaluation data (the same forecast windows used in Tables 2–3), and Appendix B then fixes these selected values as the method's hyperparameters. Thus the headline reductions in Tables 2–3 are the maxima over a p-grid on the test split, so part of the 'predicted' improvement is fitted to the evaluation window. This also makes the p<0.01 significance claim questionable, because the same data were used for selection and inference. This is not an out-of-sample prediction of the tuned configuration.

full rationale

The only formal 'first-principles' result, Proposition 1, is definitional: it states that the optimal PtC loss is no larger than the ML-only loss because the PtC policy class contains the zero-correction policy and uses the same loss. This is a property of the hypothesis class, not an empirical guarantee, and the paper itself labels it as a containment property. The empirical headline is also partly fitted: the top-p update ratio was chosen per dataset from sensitivity curves on the evaluation data (peaks at p=0.16 and p=0.10), and Appendix B then reports those selected values as the final hyperparameters, so the reported error reductions are maxima over the p-grid. Additionally, the paper claims 'all observed improvements are statistically significant at the p<0.01 level' without describing any test, but that is a statistical-validity issue rather than a circular-reasoning issue. The PtC loop itself is an independent, empirically evaluated method on external datasets and could in principle fail, so the central derivation is not equivalent to its inputs. The score reflects the vacuous theoretical theorem and the test-set-tuned hyperparameter, while recognizing that the core empirical contribution has independent content.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The framework rests on standard ML components plus a heuristic top-p masking rule; no new physical or conceptual entities are introduced. Main free parameters are hand-set or tuned per dataset, notably p.

free parameters (5)
  • top-p update ratio p = 0.10 (Walmart), 0.16 (beverage)
    Selected via sensitivity analysis (§4.3) on the same datasets used for evaluation; controls the fraction of low-magnitude parameters updated and directly affects results.
  • action clipping range [-1, 2] = [-1, 2]
    Hand-set bound on the correction factor (Eq. 2); restricts achievable corrections and influences the optimization landscape.
  • similar-SKU augmentation parameters (k, distance threshold d, minimum training size) = not reported
    Neighbor count and distance threshold for augmentation (§3.1); estimated from training data but values are not given, so re-implementation requires guessing.
  • sliding-window length = 3
    Set to 3 in Appendix A/B; determines how many recent tuples are used for online updating.
  • network hyperparameters (learning rate, hidden dim, layers, entropy) = 3e-4, 256, 3, 0.02
    Chosen by hand; standard but affect the few-shot update behavior.
assumptions (6)
  • domain assumption Immediate reward feedback is available at every time step (y_t observed before the next decision).
    Section 3.2.2 computes reward from realized demand; all online experiments assume no delayed feedback. If feedback is delayed, the loop cannot update as designed.
  • domain assumption The zero-correction policy is in the policy class (0 in A, pi_0(s)=0 in Pi).
    Used in Proposition 1; it is a design choice about the action space and makes the optimality result a containment statement.
  • ad hoc to paper Small-magnitude parameters are the safe subset to update and preserve pretrained knowledge.
    The paper explicitly calls this a heuristic ('its effectiveness must therefore be established empirically', §3.2.4). It is load-bearing for FSCCB's claimed advantage.
  • domain assumption Similar SKUs in feature space share forecast-error correction structure.
    Needed for the similar-SKU augmentation in §3.1; if feature similarity does not track error-correction similarity, augmentation may hurt.
  • standard math The smoothed objective rho_epsilon is lower bounded and L-smooth for Proposition 3.
    Standard descent-lemma conditions, invoked in Appendix F; the proof is correct under these assumptions.
  • domain assumption Demand categories and cluster-specific ML models from prior work apply to the datasets.
    The category segmentation (stable/high, stable/low, erratic/intermittent) is taken from Ma et al. (2025) and used without re-validation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Predict-then-Correct Loop Based on Few-Shot Continuous Contextual Bandit for Demand Forecasting." pith.science (2026). https://pith.science/paper/VPHVRT6B

@misc{pith2026260716354,
  author       = {Pith},
  title        = {Pith review of: A Predict-then-Correct Loop Based on Few-Shot Continuous Contextual Bandit for Demand Forecasting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VPHVRT6B}},
  note         = {Machine review of arXiv:2607.16354}
}
read the original abstract

Retail demand forecasting remains difficult when demand shifts faster than static forecasting models can be retrained, especially in early demand cycles where newly observed labels are sparse. To address this, this study aims to improve adaptive retail forecasting by proposing a predict-then-correct (PtC) framework that retains a first-stage machine learning (ML) forecast and applies a few-shot continuous contextual bandit correction policy with similar-SKUs augmentation and top-p masked updating. Across Walmart retail data and an exclusive beverage dataset, PtC delivers statistically significant reductions in MAPE, MAE, and RMSE across stable & high volume, stable & low volume, and erratic & intermittent demand patterns, improves average RMSE by 9.52% over the ML-only baseline in the ablation study, and yields lower inventory costs than base-stock, proximal policy optimization, and soft actor-critic policies under the tested lead-time settings. These findings show that online forecast correction can bridge offline demand learning and real-time retail decision-making by adapting to sparse feedback without fully retraining the base forecasting model.

Figures

Figures reproduced from arXiv: 2607.16354 by the authors.

Figure 1
Figure 1. illustrates the overall framework of the proposed PtC method. In the first stage, using average demand, coefficient of variation (CoV), and intermittency as clustering features, SKUs will be divided into four categories: stable & high volume, stable & low volume, erratic & intermittent, and lumpy (Ma et al., 2025). Each category is then associated with an ML algorithm for prediction. For the target SKU, the first st… view at source ↗
Figure 2
Figure 2. Main steps of the second stage in PtC. The bandit feedback structure is essential in this formulation. Before the actual demand yt is observed, the model only has access to st and must commit to one correction action at . After yt 11 [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Parameter sensitivity analysis of the update proportion [PITH_FULL_IMAGE:figures/full_fig_p020_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Network depth sensitivity analysis. 4.4 Ablation study To isolate the contribution of the CB correction layer and the few-shot selective update rule, we conduct an ablation study using N-BEATS as the first-stage ML forecast. We compare three variants as follows: (1) ML…
Figure 5
Figure 5. Figure 5: Inventory-cost comparison of BS, PPO, SAC and PtC under different lead times. [PITH_FULL_IMAGE:figures/full_fig_p024_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

155 extracted references · 5 linked inside Pith

  1. [1]

    Stronger forecasting in operations management—even with weak data , year =

  2. [2]

    International Journal of Production Economics , volume=

    Forecasting of customer demands for production planning by local k-nearest neighbor models , author=. International Journal of Production Economics , volume=. 2021 , publisher=

    2021

  3. [3]

    International Journal of Forecasting , volume=

    Service-level anchoring in demand forecasting: The moderating impact of retail promotions and product perishability , author=. International Journal of Forecasting , volume=. 2025 , publisher=

    2025

  4. [4]

    Transportation Research Part E: Logistics and Transportation Review , volume=

    Real-time demand forecasting for an urban delivery platform , author=. Transportation Research Part E: Logistics and Transportation Review , volume=. 2021 , publisher=

    2021

  5. [5]

    Computers and Electrical Engineering , volume=

    Real-time deep learning-based market demand forecasting and monitoring , author=. Computers and Electrical Engineering , volume=. 2022 , publisher=

    2022

  6. [6]

    Knowledge-Based Systems , volume=

    Predictive analytics for demand forecasting: A deep learning-based decision support system , author=. Knowledge-Based Systems , volume=. 2022 , publisher=

    2022

  7. [7]

    Journal of Hydrology , volume=

    A novel deep neural network architecture for real-time water demand forecasting , author=. Journal of Hydrology , volume=. 2021 , publisher=

    2021

  8. [8]

    Expert Systems with Applications , volume=

    Predicting demand for new products in fashion retailing using censored data , author=. Expert Systems with Applications , volume=. 2025 , publisher=

    2025

Show all 155 references
  1. [9]

    Computers & Industrial Engineering , volume=

    New product demand forecasting and production capacity adjustment strategies: Within-product and cross-product word-of-mouth , author=. Computers & Industrial Engineering , volume=. 2023 , publisher=

    2023

  2. [10]

    Intelligent Systems with Applications , pages=

    A multi-source multi-layer-based transfer learning approach for forecasting customer demands of newly launched products , author=. Intelligent Systems with Applications , pages=. 2025 , publisher=

    2025

  3. [11]

    Neural Computation , volume=

    Long short-term memory , author=. Neural Computation , volume=. 1997 , publisher=

    1997

  4. [12]

    Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining , pages=

    Xgboost: A scalable tree boosting system , author=. Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining , pages=

  5. [13]

    Annals of Operations Research , volume=

    Intermittent demand forecasting with transformer neural networks , author=. Annals of Operations Research , volume=. 2024 , publisher=

    2024

  6. [14]

    The Annals of Statistics , volume =

    Greedy Function Approximation: A Gradient Boosting Machine , author =. The Annals of Statistics , volume =

  7. [15]

    Annals of Tourism Research , volume=

    Enhancing tourism demand forecasting with a transformer-based framework , author=. Annals of Tourism Research , volume=. 2024 , publisher=

    2024

  8. [16]

    Manufacturing & Service Operations Management , year=

    Pooling and boosting for demand prediction in retail: A transfer learning approach , author=. Manufacturing & Service Operations Management , year=

  9. [17]

    International Journal of Forecasting , volume=

    M5 accuracy competition: Results, findings, and conclusions , author=. International Journal of Forecasting , volume=. 2022 , publisher=

    2022

  10. [18]

    International Journal of Forecasting , volume=

    Forecasting: theory and practice , author=. International Journal of Forecasting , volume=. 2022 , publisher=

    2022

  11. [19]

    Energy , volume=

    A reinforcement learning-based online learning strategy for real-time short-term load forecasting , author=. Energy , volume=. 2024 , publisher=

    2024

  12. [20]

    Advances in Neural Information Processing Systems , volume=

    Learning both weights and connections for efficient neural network , author=. Advances in Neural Information Processing Systems , volume=

  13. [21]

    Nico and Disney, Stephen M

    Boute, R. Nico and Disney, Stephen M. and Lambrecht, Marc R. and Mantin, Benny and Gijsbrechts, Jeroen , title =. European Journal of Operational Research , volume =

  14. [22]

    Manufacturing & Service Operations Management , volume=

    Can deep reinforcement learning improve inventory management? Performance on lost sales, dual-sourcing, and multi-echelon problems , author=. Manufacturing & Service Operations Management , volume=. 2022 , publisher=

    2022

  15. [23]

    2022 , publisher=

    Demand prediction in retail: A practical guide to leverage data and predictive analytics , author=. 2022 , publisher=

    2022

  16. [24]

    European Journal of Operational Research , volume=

    Forecasting daily supermarket sales using exponentially weighted quantile regression , author=. European Journal of Operational Research , volume=. 2007 , publisher=

    2007

  17. [25]

    Journal of the Operational Research Society , volume=

    Multi-item sales forecasting with total and split exponential smoothing , author=. Journal of the Operational Research Society , volume=. 2011 , publisher=

    2011

  18. [26]

    European Journal of Operational Research , volume=

    Demand forecasting with high dimensional data: The case of SKU retail sales forecasting with intra-and inter-category promotional information , author=. European Journal of Operational Research , volume=. 2016 , publisher=

    2016

  19. [27]

    European Journal of Operational Research , volume=

    The value of competitive information in forecasting FMCG retail product sales and the variable selection problem , author=. European Journal of Operational Research , volume=. 2014 , publisher=

    2014

  20. [28]

    European Journal of Operational Research , volume=

    Forecasting retailer product sales in the presence of structural change , author=. European Journal of Operational Research , volume=. 2019 , publisher=

    2019

  21. [29]

    International Journal of Production Economics , volume=

    Demand forecasting in the presence of systematic events: Cases in capturing sales promotions , author=. International Journal of Production Economics , volume=. 2020 , publisher=

    2020

  22. [30]

    International Journal of Production Economics , volume=

    Analysis and forecasting of demand during promotions for perishable items , author=. International Journal of Production Economics , volume=. 2016 , publisher=

    2016

  23. [31]

    International Journal of Production Economics , volume=

    A hybrid seasonal autoregressive integrated moving average and quantile regression for daily food sales forecasting , author=. International Journal of Production Economics , volume=. 2015 , publisher=

    2015

  24. [32]

    International Journal of Forecasting , volume=

    Forecasting retail sales using alternative models , author=. International Journal of Forecasting , volume=. 1986 , publisher=

    1986

  25. [33]

    International Journal of Forecasting , volume=

    Forecasting seasonal demand for retail: A Fourier time-varying grey model , author=. International Journal of Forecasting , volume=. 2024 , publisher=

    2024

  26. [34]

    International Journal of Research in Marketing , volume=

    Estimating the SCAN⁎ PRO model of store sales: HB, FM or just OLS? , author=. International Journal of Research in Marketing , volume=. 2008 , publisher=

    2008

  27. [35]

    Journal of Retailing and Consumer Services , volume=

    Forecasting aggregate retail sales:: a comparison of artificial neural networks and traditional methods , author=. Journal of Retailing and Consumer Services , volume=. 2001 , publisher=

    2001

  28. [36]

    Journal of Forecasting , volume=

    How effective are neural networks at forecasting and prediction? A review and evaluation , author=. Journal of Forecasting , volume=. 1998 , publisher=

    1998

  29. [37]

    Applied Soft Computing , volume=

    Improved supply chain management based on hybrid demand forecasts , author=. Applied Soft Computing , volume=. 2007 , publisher=

    2007

  30. [38]

    Transportmetrica A: Transport Science , volume=

    Short-term traffic flow prediction based on spatio-temporal analysis and CNN deep learning , author=. Transportmetrica A: Transport Science , volume=. 2019 , publisher=

    2019

  31. [39]

    International Journal of Production Research , volume=

    Sales forecasting of a food and beverage company using deep clustering frameworks , author=. International Journal of Production Research , volume=. 2024 , publisher=

    2024

  32. [40]

    Neural Networks , volume=

    Framewise phoneme classification with bidirectional LSTM and other neural network architectures , author=. Neural Networks , volume=. 2005 , publisher=

    2005

  33. [41]

    Expert Systems with Applications , volume=

    Improving flight delays prediction by developing attention-based bidirectional LSTM network , author=. Expert Systems with Applications , volume=. 2024 , publisher=

    2024

  34. [42]

    Expert Systems with Applications , volume=

    Adversarial self-attentive time-variant neural networks for multi-step time series forecasting , author=. Expert Systems with Applications , volume=. 2023 , publisher=

    2023

  35. [43]

    Production and Operations Management , volume =

    Liu, Xiaoxu and Hu, Meng and Peng, Yong and Yang, Ying , title =. Production and Operations Management , volume =

  36. [44]

    Expert Systems with Applications , volume=

    Inventory management of new products in retailers using model-based deep reinforcement learning , author=. Expert Systems with Applications , volume=. 2023 , publisher=

    2023

  37. [45]

    Expert Systems with Applications , volume=

    Deep reinforcement learning approach for solving joint pricing and inventory problem with reference price effects , author=. Expert Systems with Applications , volume=. 2022 , publisher=

    2022

  38. [46]

    and Ho, Teck-Hua and Ren, Z

    Cohen, Morris A. and Ho, Teck-Hua and Ren, Z. John and Terwiesch, Christian , title =. Management Science , volume =

  39. [47]

    John and Ho, Teck-Hua and Cohen, Morris A

    Terwiesch, Christian and Ren, Z. John and Ho, Teck-Hua and Cohen, Morris A. , title =. Management Science , volume =

  40. [48]

    International Journal of Production Research , volume =

    Chiang, Der-Jiunn and Guo, Ruey-Shun and Chen, An and Cheng, Ming-Tsang and Chen, Chiu-Bin , title =. International Journal of Production Research , volume =

  41. [49]

    , title =

    Wang, Wei and Rivera, Daniel E. , title =. IEEE Transactions on Control Systems Technology , volume =

  42. [50]

    IEEE Transactions on Systems, Man, and Cybernetics: Systems , volume =

    Chen, Ming-Chuan and Hsiao, Yu-Hsiu and Huang, Hsuan-Yu , title =. IEEE Transactions on Systems, Man, and Cybernetics: Systems , volume =

  43. [51]

    and Hong, Seung Ho , title =

    Lee, Yoon Hee and Lee, Hau L. and Hong, Seung Ho , title =. Production Planning & Control , volume =

  44. [52]

    , title =

    Mallik, Suman and Harker, Patrick T. , title =. European Journal of Operational Research , volume =

  45. [53]

    Proceedings of the National Academy of Sciences , volume=

    Overcoming catastrophic forgetting in neural networks , author=. Proceedings of the National Academy of Sciences , volume=. 2017 , publisher=

    2017

  46. [54]

    Neural Networks , volume=

    A survey on few-shot class-incremental learning , author=. Neural Networks , volume=. 2024 , publisher=

    2024

  47. [55]

    arXiv Preprint arXiv:2502.06042 , year=

    Scaling laws for forgetting during finetuning with pretraining data injection , author=. arXiv Preprint arXiv:2502.06042 , year=

    arXiv

  48. [56]

    Advances in Neural Information Processing Systems , volume=

    Movement pruning: Adaptive sparsity by fine-tuning , author=. Advances in Neural Information Processing Systems , volume=

  49. [57]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Slca: Slow learner with classifier alignment for continual learning on a pre-trained model , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  50. [58]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Few-shot class-incremental learning , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  51. [59]

    arXiv Preprint arXiv:2003.00075 , year=

    Learned threshold pruning , author=. arXiv Preprint arXiv:2003.00075 , year=

    2003 arXiv

  52. [60]

    European conference on computer vision , pages=

    Filter pruning via feature discrimination in deep neural networks , author=. European conference on computer vision , pages=. 2022 , organization=

    2022

  53. [61]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Few-shot lifelong learning , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  54. [62]

    Uzsoy, Reha and Fowler, John W. and M. A survey of semiconductor supply chain models part II: demand planning, inventory management, and capacity planning , journal =

  55. [63]

    Computers & Industrial Engineering , pages =

    Kuo, Hsiao-An and Hong, Ting-Yun and Chien, Chih-Fang , title =. Computers & Industrial Engineering , pages =

  56. [64]

    Advances in Neural Information Processing Systems , volume=

    Actor-critic algorithms , author=. Advances in Neural Information Processing Systems , volume=

  57. [65]

    arXiv Preprint arXiv:1707.06347 , year=

    Proximal policy optimization algorithms , author=. arXiv Preprint arXiv:1707.06347 , year=

    arXiv

  58. [66]

    International conference on machine learning , pages=

    Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor , author=. International conference on machine learning , pages=. 2018 , organization=

    2018

  59. [67]

    Industrial & Engineering Chemistry Research , volume=

    Soft actor-critic deep reinforcement learning with hybrid mixed-integer actions for demand responsive scheduling of energy systems , author=. Industrial & Engineering Chemistry Research , volume=. 2022 , publisher=

    2022

  60. [68]

    Energy , volume=

    High robustness energy management strategy of hybrid electric vehicle based on improved soft actor-critic deep reinforcement learning , author=. Energy , volume=. 2022 , publisher=

    2022

  61. [69]

    and Padmanabhan, V

    Lee, Hau L. and Padmanabhan, V. and Whang, Seungjin , title =. Sloan Management Review , volume =

  62. [70]

    , title =

    Wang, Xiaohang and Disney, Stephen M. , title =. European Journal of Operational Research , volume =

  63. [71]

    International Journal of Production Economics , volume =

    Cao, Chen and He, Yu and Liu, Yang and Huang, Hao and Zhang, Fei , title =. International Journal of Production Economics , volume =

  64. [72]

    IEEE Transactions on Evolutionary Computation , volume =

    Chen, Long and Liu, Hao-Lun and Tan, Kay Chen and Li, Ke , title =. IEEE Transactions on Evolutionary Computation , volume =

  65. [73]

    International Journal of Production Economics , volume =

    Chien, Chih-Fang and Chen, Yueh-Ju and Peng, Jia-Tzer , title =. International Journal of Production Economics , volume =

  66. [74]

    International Journal of Production Research , volume =

    Chien, Chih-Fang and Lin, Yen-Sheng and Lin, Shi-Kai , title =. International Journal of Production Research , volume =

  67. [75]

    2018 , publisher=

    Forecasting: principles and practice , author=. 2018 , publisher=

    2018

  68. [76]

    Engineering Applications of Artificial Intelligence , volume=

    A hybrid ETS--ANN model for time series forecasting , author=. Engineering Applications of Artificial Intelligence , volume=. 2017 , publisher=

    2017

  69. [77]

    International Journal of Logistics Research and Applications , pages =

    Ma, Benedict Jun and Jackson, Isaac and Huang, Min and Villegas, Sergio and Macias-Aguayo, Jose , title =. International Journal of Logistics Research and Applications , pages =

  70. [78]

    and Boylan, John E

    Syntetos, Aris A. and Boylan, John E. and Croston, John D. , title =. Journal of the Operational Research Society , volume =

  71. [79]

    Journal of Machine Learning Research , volume=

    River: machine learning for streaming data in python , author=. Journal of Machine Learning Research , volume=

  72. [80]

    Advances in Neural Information Processing Systems , volume=

    Lightgbm: A highly efficient gradient boosting decision tree , author=. Advances in Neural Information Processing Systems , volume=

  73. [81]

    Journal of Forecasting , volume=

    Exponential smoothing: The state of the art , author=. Journal of Forecasting , volume=. 1985 , publisher=

    1985

  74. [82]

    arXiv Preprint arXiv:1905.10437 , year=

    N-BEATS: Neural basis expansion analysis for interpretable time series forecasting , author=. arXiv Preprint arXiv:1905.10437 , year=

    1905 arXiv

  75. [83]

    arXiv Preprint arXiv:2304.08424 , year=

    Long-term Forecasting with TiDE: Time-series Dense Encoder , author=. arXiv Preprint arXiv:2304.08424 , year=

    arXiv

  76. [84]

    2025 IEEE/ACIS 23rd International Conference on Software Engineering Research, Management and Applications (SERA) , pages=

    Automated Demand Forecasting in Retail Supply Chains Using Deep Reinforcement Learning , author=. 2025 IEEE/ACIS 23rd International Conference on Software Engineering Research, Management and Applications (SERA) , pages=. 2025 , organization=

    2025

  77. [85]

    International Journal of Forecasting , volume=

    Temporal fusion transformers for interpretable multi-horizon time series forecasting , author=. International Journal of Forecasting , volume=. 2021 , publisher=

    2021

  78. [86]

    International conference on machine learning , pages=

    Asynchronous methods for deep reinforcement learning , author=. International conference on machine learning , pages=. 2016 , organization=

    2016

  79. [87]

    SoftwareX , volume=

    TfELM: Extreme learning machines framework with python and TensorFlow , author=. SoftwareX , volume=. 2024 , publisher=

    2024

  80. [88]

    Knowledge-Based Systems , volume =

    Zhang, Hao and Li, Jing and Zhang, Jian and Dong, Ying , title =. Knowledge-Based Systems , volume =

  81. [89]

    Pattern Recognition , volume =

    Zhang, Ziyang and Chen, Xin and Wang, Chen and Wang, Rui and Song, Wei and Nie, Feiping , title =. Pattern Recognition , volume =

  82. [90]

    Expert Systems with Applications , volume =

    Huber, Johannes and Gossmann, Anja and Stuckenschmidt, Heiner , title =. Expert Systems with Applications , volume =

  83. [91]

    Information Processing & Management , volume =

    Kaya, Kaan and Yılmaz, Yusuf and Yaslan, Yusuf and Öğüdücü, Şule Gündüz and Çıngı, Fatma , title =. Information Processing & Management , volume =

  84. [92]

    International Journal of Logistics Research and Applications , volume =

    Chien, Chih-Fang and Kuo, Hsiao-An and Lin, Yen-Sheng , title =. International Journal of Logistics Research and Applications , volume =

  85. [93]

    International Journal of Engineering Research in Africa , volume =

    El Jaouhari, Ayoub and Alhilali, Zakaria and Arif, Jamal and Fellaki, Said and Amejwal, Mohammed and Azzouz, Khalid , title =. International Journal of Engineering Research in Africa , volume =

  86. [94]

    International Journal of Production Economics , volume =

    Fan, Shi-Kai and Chiu, Shu-Hua , title =. International Journal of Production Economics , volume =

  87. [95]

    Production Planning & Control , volume =

    Gallego, Guillermo and Katircioglu, Kivanc and Ramachandran, Bala , title =. Production Planning & Control , volume =

  88. [96]

    Proceedings of the 2006 IEEE International Symposium on Semiconductor Manufacturing , pages =

    Guo, Ruey-Shun and Chiang, Der-Jiunn and Yang, Ping-Chang , title =. Proceedings of the 2006 IEEE International Symposium on Semiconductor Manufacturing , pages =

    2006

  89. [97]

    An agent-based infrastructure for assessing the performance of planning approaches for semiconductor supply chains , journal =

    Herding, Ramon and M. An agent-based infrastructure for assessing the performance of planning approaches for semiconductor supply chains , journal =

  90. [98]

    Neural Computing and Applications , volume =

    Hirtz, Tobias and Tian, Hao and Shahzad, Shaharyar and Wu, Fan and Yang, Yanfeng and Ren, Tian-Ling , title =. Neural Computing and Applications , volume =

  91. [99]

    Simulation Modelling Practice and Theory , volume =

    Kim, Dong and Park, Young Soo and Kim, Hohyun and Park, Kyung Soo and Moon, Il , title =. Simulation Modelling Practice and Theory , volume =

  92. [100]

    Environmental Science & Technology , volume =

    Krishnan, Nikhil and Boyd, Stephen and Somani, Arun and Raoux, Sylvain and Clark, Daniel and Dornfeld, David , title =. Environmental Science & Technology , volume =

  93. [101]

    Expert Systems with Applications , volume =

    Lee, Young Hae and Lee, Seung , title =. Expert Systems with Applications , volume =

  94. [102]

    IISE Transactions , volume =

    Li, Yutong and Du, Jun and Jiang, Wen , title =. IISE Transactions , volume =

  95. [103]

    Journal Europ

    Li, Bo , title =. Journal Europ

  96. [104]

    Hasnat , title =

    Liao, Zhen and Tantai, Bowen and Abdul-Hamid, Ahmad-Qamar and Mukhtar, Danish and Ali, M. Hasnat , title =. International Journal of Production Economics , pages =

  97. [105]

    International Review of Economics & Finance , volume =

    Lien, Dar-Yuan and Wang, Li-Hsien , title =. International Review of Economics & Finance , volume =

  98. [106]

    Economic Modelling , volume =

    Liu, Wen-Hsiung and Chung, Chia-Feng and Chang, Kuan-Lin , title =. Economic Modelling , volume =

  99. [107]

    Complex & Intelligent Systems , volume =

    Liu, Jie and Qiao, Feng and Zou, Meng and Zinn, John and Ma, Yanna and Vogel-Heuser, Birgit , title =. Complex & Intelligent Systems , volume =

  100. [108]

    IEEE Transactions on Semiconductor Manufacturing , year =

    Liu, Boyang and Zhao, Dong and Liu, Yang and Lu, Xin , title =. IEEE Transactions on Semiconductor Manufacturing , year =

  101. [109]

    IEEE Transactions on Automation Science and Engineering , volume =

    Park, In-Bum and Huh, Joon and Kim, Jongkeun and Park, Jinwoo , title =. IEEE Transactions on Automation Science and Engineering , volume =

  102. [110]

    IEEE Transactions on Cybernetics , volume =

    Park, In-Bum and Park, Jinwoo , title =. IEEE Transactions on Cybernetics , volume =

  103. [111]

    Vijay, Yamini , title =

  104. [112]

    Computers & Industrial Engineering , volume =

    Wang, Xin and Zhan, Li and Zhang, Ying and Fei, Ting and Tseng, Ming-Lang , title =. Computers & Industrial Engineering , volume =

  105. [113]

    IISE Transactions , volume=

    Robust purchasing and information asymmetry in supply chains with a price-only contract , author=. IISE Transactions , volume=. 2015 , publisher=

    2015

  106. [114]

    IISE Transactions , volume=

    Supply chain hoarding and contingent sourcing strategies in anticipation of price hikes and product shortages , author=. IISE Transactions , volume=. 2024 , publisher=

    2024

  107. [115]

    IISE Transactions , volume=

    Early sale of seasonal inventory in the newsvendor problem , author=. IISE Transactions , volume=. 2019 , publisher=

    2019

  108. [116]

    IISE Transactions , volume=

    Forecasting and information sharing in supply chains under ARMA demand , author=. IISE Transactions , volume=. 2014 , publisher=

    2014

  109. [117]

    IISE Transactions , volume=

    A multi-agent reinforcement learning model for inventory transshipments under supply chain disruption , author=. IISE Transactions , volume=. 2024 , publisher=

    2024

  110. [118]

    IISE Transactions , volume=

    Dynamic pricing of regulated field services using reinforcement learning , author=. IISE Transactions , volume=. 2023 , publisher=

    2023

  111. [119]

    IISE Transactions , volume=

    Game-theoretic analyses of strategic pricing decision problems in supply chains , author=. IISE Transactions , volume=. 2020 , publisher=

    2020

  112. [120]

    International Journal of Forecasting , volume=

    Exponential smoothing with a damped multiplicative trend , author=. International Journal of Forecasting , volume=

  113. [121]

    and Boylan, John E

    Nikolopoulos, Konstantinos and Syntetos, Aris A. and Boylan, John E. and Petropoulos, Fotios and Assimakopoulos, Vassilios , journal=. An aggregate-disaggregate intermittent demand approach (

  114. [122]

    Journal of the Operational Research Society , volume=

    Forecasting and operational research: A review , author=. Journal of the Operational Research Society , volume=

  115. [123]

    International Journal of Forecasting , volume=

    Analysis of judgmental adjustments in the presence of promotions , author=. International Journal of Forecasting , volume=

  116. [124]

    Journal of the Operational Research Society , volume=

    On the identification of sales forecasting models in the presence of promotions , author=. Journal of the Operational Research Society , volume=

  117. [125]

    Journal of Business Research , volume=

    Simple versus complex forecasting: The evidence , author=. Journal of Business Research , volume=

  118. [126]

    Journal of Forecasting , volume=

    The accuracy of extrapolation (time series) methods: Results of a forecasting competition , author=. Journal of Forecasting , volume=

  119. [127]

    Makridakis, Spyros and Hibon, Michele and Lusk, Edward and Belhadjali, Mohamed , journal=. The

  120. [128]

    Makridakis, Spyros and Hibon, Michele , journal=. The

  121. [129]

    Operations Research , volume=

    Inventory management of a fast-fashion retail network , author=. Operations Research , volume=

  122. [130]

    Production and Operations Management , volume=

    The value of weather information for e-commerce operations , author=. Production and Operations Management , volume=

  123. [131]

    and Sanders, Nada R

    Boone, Tonya and Ganeshan, Ram and Hicks, Robert L. and Sanders, Nada R. , journal=. Can

  124. [132]

    Production and Operations Management , volume=

    The operational value of social media information , author=. Production and Operations Management , volume=

  125. [133]

    International Journal of Forecasting , volume=

    Retail forecasting: Research and practice , author=. International Journal of Forecasting , volume=

  126. [134]

    and Qi, Y

    Lei, Dong and Hu, Huifeng and Geng, Dong and Zhang, J. and Qi, Y. and Liu, S. and Shen, Zuo-Jun Max , journal=. New product life cycle curve modeling and forecasting with product attributes and promotion: A

  127. [135]

    Manufacturing & Service Operations Management , volume=

    Analytics for an online retailer: Demand forecasting and price optimization , author=. Manufacturing & Service Operations Management , volume=

  128. [136]

    Manufacturing & Service Operations Management , volume=

    Pricing for heterogeneous products: Analytics for ticket reselling , author=. Manufacturing & Service Operations Management , volume=

  129. [137]

    Manufacturing & Service Operations Management , volume=

    Bayesian ensembles of exponentially smoothed life-cycle forecasts , author=. Manufacturing & Service Operations Management , volume=

  130. [138]

    Operations and Supply Chain Management: An International Journal , volume=

    Data analytics in the supply chain management: Review of machine learning applications in demand forecasting , author=. Operations and Supply Chain Management: An International Journal , volume=

  131. [139]

    International Journal of Logistics Research and Applications , volume=

    Machine learning demand forecasting and supply chain performance , author=. International Journal of Logistics Research and Applications , volume=

  132. [140]

    2021 12th International Conference on Computing Communication and Networking Technologies , pages=

    Demand forecasting: Literature review on various methodologies , author=. 2021 12th International Conference on Computing Communication and Networking Technologies , pages=

    2021

  133. [141]

    International Journal of Production Economics , volume=

    Machine learning and optimization models for supplier selection and order allocation planning , author=. International Journal of Production Economics , volume=

  134. [142]

    Journal of Integrated Science and Technology , volume=

    A systematic study on effective demand prediction using machine learning , author=. Journal of Integrated Science and Technology , volume=

  135. [143]

    Computers and Electronics in Agriculture , volume=

    Corn cash price forecasting with neural networks , author=. Computers and Electronics in Agriculture , volume=

  136. [144]

    Journal of Uncertain Systems , volume=

    Machine learning coffee price predictions , author=. Journal of Uncertain Systems , volume=

  137. [145]

    and Zhang, Y

    Xu, X. and Zhang, Y. , journal=. Price forecasts of ten steel products using

  138. [146]

    and Xu, X

    Jin, B. and Xu, X. , journal=

  139. [147]

    and Xu, X

    Jin, B. and Xu, X. , journal=. Forecasts of coking coal futures price indices through

  140. [148]

    and Xu, X

    Jin, B. and Xu, X. , journal=. Machine learning-based forecasts of residential property prices in

  141. [149]

    and Xu, X

    Jin, B. and Xu, X. , journal=. High-frequency

  142. [150]

    , journal=

    Xu, X. , journal=. Contemporaneous and

  143. [151]

    Decision Analytics Journal , volume=

    An integrated vector error correction and directed acyclic graph method for investigating contemporaneous causalities , author=. Decision Analytics Journal , volume=

  144. [152]

    and Xu, X

    Jin, B. and Xu, X. , journal=. A study of contemporaneous residential real estate price causation across major

  145. [153]

    Journal of Applied Statistics , volume=

    Short-run price forecast performance of individual and composite models for 496 corn cash markets , author=. Journal of Applied Statistics , volume=

  146. [154]

    American Journal of Agricultural Economics , volume=

    Corn cash price forecasting , author=. American Journal of Agricultural Economics , volume=

  147. [155]

    and Zhang, Y

    Xu, X. and Zhang, Y. , journal=. Individual time series and composite forecasting of the

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.