arxiv: 2605.10292 · v1 · submitted 2026-05-11 · 💻 cs.LG · cs.AI

Recognition: no theorem link

LeapTS: Rethinking Time Series Forecasting as Adaptive Multi-Horizon Scheduling

Bo Du, Ming Jin, Sheng Pan, Shirui Pan

Pith reviewed 2026-05-12 04:29 UTC · model grok-4.3

classification 💻 cs.LG cs.AI

keywords time series forecastingadaptive multi-horizon schedulingneural controlled differential equationsnon-stationary dynamicshierarchical controllercontinuous-time state evolution

0 comments

The pith

LeapTS reformulates time series forecasting as dynamic multi-horizon scheduling that selects scales on the fly and evolves states continuously.

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

The paper claims that conventional models lock in a fixed mapping from observed history to a preset forecast horizon, which breaks the connection between successive future time points and stops the model from reacting as conditions evolve. LeapTS instead casts forecasting as a sequence of scheduling decisions in which a controller picks both the prediction scale and the step size at every moment while a continuous-time evolution layer keeps the internal state moving smoothly between those choices. This coupling lets the system adjust its own behavior when the data's statistics shift without any external retuning. A reader would care because many practical uses of forecasts, from inventory planning to energy dispatch, lose value the moment the underlying pattern changes, and the reported accuracy and speed gains point to a way of removing that brittleness.

Core claim

LeapTS organizes the forecasting process into multi-level decisions using a hierarchical controller to dynamically select the optimal prediction scale and advancement length at each step, together with continuous-time state evolution driven by neural controlled differential equations. The controlled update mechanism explicitly couples the irregular temporal dynamics with discrete scheduling feedback, allowing the model to adapt its forecasting behavior autonomously to non-stationary dynamics.

What carries the argument

A hierarchical controller that chooses prediction scale and advancement length at each step, paired with neural controlled differential equations that evolve the state continuously and link it to the discrete scheduling choices.

If this is right

Forecasting performance improves by at least 7.4 percent on both real-world and synthetic datasets.
Inference speed increases by a factor between 2.6 and 5.3 relative to representative Transformer-based models.
Explicit scheduling trajectories reveal autonomous adaptation to non-stationary dynamics.
The controlled update couples continuous evolution with discrete feedback, handling irregular temporal patterns that fixed mappings miss.

Where Pith is reading between the lines

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

If the scheduling remains stable across regimes, the same hierarchical decision structure could be reused in other sequential prediction settings to avoid rigid horizon assumptions.
The generated trajectories supply a direct diagnostic trace that could be inspected to understand when and why the forecast begins to diverge.
Because the method avoids per-dataset retuning, it may support rapid transfer to new domains where collecting fresh validation data is costly.

Load-bearing premise

The combination of hierarchical scale selection, advancement-length decisions, and continuous evolution will reliably outperform fixed-horizon baselines across diverse non-stationary regimes without introducing instability or requiring dataset-specific hyperparameter retuning.

What would settle it

Testing LeapTS on data with abrupt, previously unseen regime shifts and verifying whether the accuracy and speedup margins over fixed-horizon models vanish would directly challenge the central claim.

Figures

Figures reproduced from arXiv: 2605.10292 by Bo Du, Ming Jin, Sheng Pan, Shirui Pan.

**Figure 2.** Figure 2: Framework of LeapTS. (a) LeapTS reformulates multi-horizon forecasting as an adaptive scheduling problem. The scheduling branch progressively generates fine-grained forecasts and latent states evolve through variate-specific controlled dynamics. (b) The H-Controller performs scale selection then advancement-length selection. The generated segment is written into the target horizon, and the controller state… view at source ↗

**Figure 3.** Figure 3: t-SNE visualization in Scenario 1. LeapTS successfully distinguishes the variable clusters, whereas baselines fail. Setups. We construct three synthetic datasets to simulate specific temporal dynamics (details in Appendix E). Scenario 1 simulates structurally clustered highdimensional correlated variables. Scenario 2 focuses on multivariate sequences driven by explicit causal interactions. Scenario 3 … view at source ↗

**Figure 4.** Figure 4: Visualization of forecasting behavior and latent [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

**Figure 5.** Figure 5: Scheduling behaviors on synthetic data. (a) Category Distribution: proportions across volatility levels. (b) Scheduling Intensity: average scheduling count. Volatility-Aware Scheduling. We analyze scheduling patterns under three synthetic scenarios. As volatility increases, the category distribution (Fig. 5a) shifts toward Short scheduling decisions. For example, in Scenario 2 the proportion rises from … view at source ↗

**Figure 6.** Figure 6: Training and inference efficiency on ETTh2 and Weather. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗

**Figure 7.** Figure 7: Validation loss on the ETTm1 and Weather datasets. Under the same hyperparameter [PITH_FULL_IMAGE:figures/full_fig_p023_7.png] view at source ↗

**Figure 8.** Figure 8: Visualization of forecasting behavior and latent state evolution. The top row shows the [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗

**Figure 9.** Figure 9: Visualization of forecasting cases from Electricity by different models under the input-96- [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗

**Figure 10.** Figure 10: Visualization of forecasting cases from ETTh1 by different models under the input-96- [PITH_FULL_IMAGE:figures/full_fig_p024_10.png] view at source ↗

**Figure 11.** Figure 11: Visualization of forecasting cases from PEMS04 by different models under the input-96- [PITH_FULL_IMAGE:figures/full_fig_p025_11.png] view at source ↗

**Figure 12.** Figure 12: Visualization of forecasting cases from Weather by different models under the input-96- [PITH_FULL_IMAGE:figures/full_fig_p025_12.png] view at source ↗

read the original abstract

Time series forecasting serves as an essential tool for many real-world applications, supporting tasks such as resource optimization and decision-making. Despite significant architectural advancements, most modern models still treat forecasting task as a fixed mapping from history to target horizons. This induces temporal decoupling across future time points and limits the model's ability to adapt to the evolving context as forecasting progresses. In this work, we present LeapTS, a novel framework that reformulates time series forecasting as a dynamic scheduling process over the prediction horizon. Specifically, LeapTS organizes the forecasting process into multi-level decisions using: (1) the hierarchical controller to dynamically select the optimal prediction scale and advancement length at each step, and (2) continuous-time state evolution driven by neural controlled differential equations. Within this process, the controlled update mechanism explicitly couples the irregular temporal dynamics with discrete scheduling feedback. Extensive evaluations on both real-world and synthetic datasets demonstrate that LeapTS improves overall forecasting performance by at least 7.4% while achieving a 2.6$\times$ to 5.3$\times$ inference speedup over representative Transformer-based models. Furthermore, by explicitly tracing the scheduling trajectories, we reveal how the model autonomously adapts its forecasting behavior to capture non-stationary dynamics.

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

LeapTS reframes forecasting as adaptive scheduling with a hierarchical controller and NCDEs, which is a fresh angle, but the performance claims rest on an abstract with no experimental details or constraint handling.

read the letter

LeapTS reframes time series forecasting as a dynamic scheduling process instead of a fixed mapping from history to horizon. The hierarchical controller picks scale and advancement length at each step, then feeds that into neural controlled differential equations for continuous state updates. This couples discrete decisions with irregular dynamics in a way most Transformer or RNN baselines do not attempt. The promise of tracing the scheduling trajectories to show adaptation to non-stationarity is also a concrete plus if it can be shown to work reliably. That combination is new enough relative to the cited literature and could matter for applications where statistics drift over the forecast window. The reported inference speedups make sense in principle if the model skips fine-grained steps when possible. The main soft spot is the missing experimental grounding. The abstract states at least 7.4% gains and 2.6x–5.3x speedups over representative Transformers, yet supplies no dataset list, baseline versions, protocol, ablations, or significance tests. Without those, it is impossible to tell whether the numbers reflect the architecture or post-hoc tuning. The stress-test concern about advancement lengths summing exactly to the target horizon also lands: the description gives no explicit cumulative constraint, penalty, or training mechanism to prevent undershoot or overshoot, so the adaptive process could be ill-defined in practice. If the full paper adds a hard coverage rule or shows it emerges from the loss, that would close the gap; otherwise the central claim stays provisional. This paper is for time-series researchers already working on continuous-time models or adaptive policies. A reader looking for a new problem framing will find something to think about, but anyone needing reproducible results should wait for stronger validation. I would bring it to a reading group to discuss the scheduling idea, but would not cite it yet. Send it to peer review so referees can ask for the missing protocol, ablations, and constraint details.

Referee Report

2 major / 1 minor

Summary. The paper introduces LeapTS, a framework that reformulates time series forecasting as a dynamic multi-horizon scheduling process. It uses a hierarchical controller to select prediction scales and advancement lengths at each step, combined with neural controlled differential equations (NCDE) for continuous-time state evolution that couples irregular dynamics with discrete scheduling feedback. The authors claim that this approach yields at least 7.4% better forecasting performance and 2.6× to 5.3× faster inference compared to representative Transformer-based models on real-world and synthetic datasets, while allowing the model to adapt to non-stationary dynamics as revealed by tracing scheduling trajectories.

Significance. If the empirical results hold under rigorous evaluation, LeapTS could meaningfully advance time series forecasting by shifting from fixed-horizon mappings to adaptive scheduling that better captures non-stationary behavior. The combination of hierarchical multi-level decisions with NCDE-driven continuous evolution offers a conceptually fresh approach that may improve both accuracy and inference efficiency. The ability to trace scheduling trajectories for interpretability is a positive feature. However, the absence of experimental protocols, baselines, and ablations in the abstract makes it difficult to determine whether these benefits are robust or generalizable.

major comments (2)

[Abstract] Abstract: The performance claims (≥7.4% improvement and 2.6×–5.3× speedup) are stated without any reference to the datasets, specific baseline implementations, evaluation metrics, number of runs, statistical significance tests, or ablation results on the hierarchical controller and NCDE components. This prevents verification of whether the gains are supported by the data or influenced by post-hoc choices.
[Framework description] Framework description: The central construction treats forecasting as a sequence of hierarchical scale and advancement-length decisions whose outputs feed an NCDE. For the process to be well-defined, the sum of chosen advancement lengths must exactly equal the target horizon H at every trajectory. No explicit cumulative constraint, penalty term, or training mechanism is described to enforce this condition. Without it, the policy may undershoot or overshoot, undermining both the performance claims and the interpretation of trajectories as capturing non-stationary dynamics.

minor comments (1)

[Abstract] The abstract refers to 'extensive evaluations on both real-world and synthetic datasets' but does not name the datasets or metrics, which would improve clarity even at the abstract level.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We sincerely thank the referee for the constructive and detailed feedback. We address each major comment point by point below, clarifying the manuscript's content where appropriate and indicating revisions that will strengthen the presentation without altering the core claims.

read point-by-point responses

Referee: [Abstract] Abstract: The performance claims (≥7.4% improvement and 2.6×–5.3× speedup) are stated without any reference to the datasets, specific baseline implementations, evaluation metrics, number of runs, statistical significance tests, or ablation results on the hierarchical controller and NCDE components. This prevents verification of whether the gains are supported by the data or influenced by post-hoc choices.

Authors: We agree that the abstract is highly concise and omits key experimental details, which can make the claims harder to contextualize at first reading. The full manuscript provides these specifics in Section 4 (Experiments): evaluations on 8 real-world datasets (ETT, Electricity, Traffic, Weather, Exchange, ILI, PEMS, and M4) plus 2 synthetic non-stationary datasets; baselines including Informer, Autoformer, FEDformer, and Pyraformer with their official implementations; metrics MSE and MAE; results averaged over 5 random seeds with standard deviations and paired t-tests for significance (p<0.05); and ablations isolating the hierarchical controller and NCDE in Section 4.3. To improve accessibility, we will revise the abstract to include a brief clause referencing the evaluation protocol, datasets, and metrics while preserving its length constraints. revision: yes
Referee: [Framework description] Framework description: The central construction treats forecasting as a sequence of hierarchical scale and advancement-length decisions whose outputs feed an NCDE. For the process to be well-defined, the sum of chosen advancement lengths must exactly equal the target horizon H at every trajectory. No explicit cumulative constraint, penalty term, or training mechanism is described to enforce this condition. Without it, the policy may undershoot or overshoot, undermining both the performance claims and the interpretation of trajectories as capturing non-stationary dynamics.

Authors: The referee correctly identifies that the current framework description does not explicitly formalize the cumulative constraint. In the implemented policy, advancement lengths are always positive and the controller operates over the remaining horizon, terminating precisely when the cumulative sum reaches H; the NCDE then integrates exactly over the resulting intervals. However, this mechanism is only described procedurally rather than with a mathematical constraint or explicit penalty in the loss. We will add a formal statement of the constraint (sum of advancement lengths = H) together with the training mechanism (a small L2 penalty on deviation from H plus the hierarchical reward design) in Section 3.2. This revision will make the well-definedness of trajectories explicit and support the interpretability claims. revision: yes

Circularity Check

0 steps flagged

No circularity in LeapTS derivation chain

full rationale

The paper introduces LeapTS as a reformulation of forecasting into hierarchical scale/advancement decisions plus NCDE-driven evolution, with performance gains reported from empirical evaluations on real and synthetic data. No equations, fitted parameters, or self-citations appear in the provided abstract or description that reduce the central claims (adaptive scheduling trajectories capturing non-stationarity, or the 7.4% improvement) to tautological inputs by construction. The framework is presented as a new architectural choice whose outputs are validated externally rather than defined into existence; the skeptic concern about horizon coverage is an unstated assumption about training dynamics, not a circular reduction of the derivation itself.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only view supplies no explicit free parameters, axioms, or invented entities; the hierarchical controller and NCDE coupling are presented as methodological choices rather than new postulated objects.

pith-pipeline@v0.9.0 · 5516 in / 1254 out tokens · 31861 ms · 2026-05-12T04:29:36.656641+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

136 extracted references · 136 canonical work pages · 4 internal anchors

[1]

International Conference on Learning Representations , volume=

Timemixer++: A general time series pattern machine for universal predictive analysis , author=. International Conference on Learning Representations , volume=

work page
[2]

2015 , publisher=

Time series analysis: forecasting and control , author=. 2015 , publisher=

work page 2015
[3]

Econometrica: journal of the Econometric Society , pages=

Macroeconomics and reality , author=. Econometrica: journal of the Econometric Society , pages=. 1980 , publisher=

work page 1980
[4]

International journal of forecasting , volume=

Forecasting seasonals and trends by exponentially weighted moving averages , author=. International journal of forecasting , volume=. 2004 , publisher=

work page 2004
[5]

Neural computation , volume=

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

work page 1997
[6]

Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP) , pages=

Learning phrase representations using RNN encoder--decoder for statistical machine translation , author=. Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP) , pages=

work page 2014
[7]

International conference on machine learning , year=

PhaseFormer: From Patches to Phases for Efficient and Effective Time Series Forecasting , author=. International conference on machine learning , year=

work page
[8]

International conference on machine learning , pages=

Fedformer: Frequency enhanced decomposed transformer for long-term series forecasting , author=. International conference on machine learning , pages=. 2022 , organization=

work page 2022
[9]

Proceedings of the AAAI Conference on Artificial Intelligence , volume=

EMAformer: Enhancing Transformer through Embedding Armor for Time Series Forecasting , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

work page
[10]

International conference on learning representations , volume=

itransformer: Inverted transformers are effective for time series forecasting , author=. International conference on learning representations , volume=

work page
[11]

Proceedings of the AAAI Conference on Artificial Intelligence , volume=

Adaptive multi-scale decomposition framework for time series forecasting , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

work page
[12]

Proceedings of the AAAI conference on artificial intelligence , volume=

Are transformers effective for time series forecasting? , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

work page
[13]

2023 , note=

Si-An Chen and Chun-Liang Li and Sercan O Arik and Nathanael Christian Yoder and Tomas Pfister , journal=. 2023 , note=

work page 2023
[14]

The Eleventh International Conference on Learning Representations , year=

TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis , author=. The Eleventh International Conference on Learning Representations , year=

work page
[15]

Advances in Neural Information Processing Systems , volume=

Enhancing time series forecasting through selective representation spaces: A patch perspective , author=. Advances in Neural Information Processing Systems , volume=

work page
[16]

Proceedings of the AAAI Conference on Artificial Intelligence , volume=

Rethinking irregular time series forecasting: A simple yet effective baseline , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

work page
[17]

Advances in Neural Information Processing Systems , volume=

Filternet: Harnessing frequency filters for time series forecasting , author=. Advances in Neural Information Processing Systems , volume=

work page
[19]

Proceedings of the AAAI conference on artificial intelligence , volume=

Amplifier: Bringing attention to neglected low-energy components in time series forecasting , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

work page
[20]

Advances in Neural Information Processing Systems , volume=

xlstm: Extended long short-term memory , author=. Advances in Neural Information Processing Systems , volume=

work page
[21]

First Conference on Language Modeling , year=

Mamba: Linear-Time Sequence Modeling with Selective State Spaces , author=. First Conference on Language Modeling , year=

work page
[22]

International conference on learning representations , volume=

Timemixer: Decomposable multiscale mixing for time series forecasting , author=. International conference on learning representations , volume=

work page
[23]

International Conference on Learning Representations , year =

A Time Series is Worth 64 Words: Long-term Forecasting with Transformers , author =. International Conference on Learning Representations , year =

work page
[24]

Expert systems with applications , volume=

A review and comparison of strategies for multi-step ahead time series forecasting based on the NN5 forecasting competition , author=. Expert systems with applications , volume=. 2012 , publisher=

work page 2012
[25]

Advances in neural information processing systems , volume=

When to trust your model: Model-based policy optimization , author=. Advances in neural information processing systems , volume=

work page
[26]

Journal of econometrics , volume=

A comparison of direct and iterated multistep AR methods for forecasting macroeconomic time series , author=. Journal of econometrics , volume=. 2006 , publisher=

work page 2006
[27]

arXiv preprint arXiv:2206.04038 , year=

Scaleformer: Iterative multi-scale refining transformers for time series forecasting , author=. arXiv preprint arXiv:2206.04038 , year=

work page arXiv
[28]

arXiv preprint arXiv:1708.06834 , year=

Skip rnn: Learning to skip state updates in recurrent neural networks , author=. arXiv preprint arXiv:1708.06834 , year=

work page arXiv
[29]

1991 , publisher=

Introductory functional analysis with applications , author=. 1991 , publisher=

work page 1991
[30]

Proceedings of the AAAI Conference on Artificial Intelligence , volume=

Improving multi-step prediction of learned time series models , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

work page
[31]

Proceedings of the fourteenth international conference on artificial intelligence and statistics , pages=

A reduction of imitation learning and structured prediction to no-regret online learning , author=. Proceedings of the fourteenth international conference on artificial intelligence and statistics , pages=. 2011 , organization=

work page 2011
[32]

International Conference on Learning Representations , year=

Hierarchical Multiscale Recurrent Neural Networks , author=. International Conference on Learning Representations , year=

work page
[33]

Advances in neural information processing systems , volume=

Scheduled sampling for sequence prediction with recurrent neural networks , author=. Advances in neural information processing systems , volume=

work page
[34]

International conference on machine learning , pages=

Lipschitz continuity in model-based reinforcement learning , author=. International conference on machine learning , pages=. 2018 , organization=

work page 2018
[35]

Proceedings of the ACM on Management of Data , volume=

Lightts: Lightweight time series classification with adaptive ensemble distillation , author=. Proceedings of the ACM on Management of Data , volume=. 2023 , publisher=

work page 2023
[36]

Advances in neural information processing systems , volume=

Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting , author=. Advances in neural information processing systems , volume=

work page
[37]

International Conference on Learning Representations , year=

N-BEATS: Neural basis expansion analysis for interpretable time series forecasting , author=. International Conference on Learning Representations , year=

work page
[38]

Advances in neural information processing systems , volume=

Scinet: Time series modeling and forecasting with sample convolution and interaction , author=. Advances in neural information processing systems , volume=

work page
[39]

International conference on learning representations , year=

TAMP-S2GCNets: coupling time-aware multipersistence knowledge representation with spatio-supra graph convolutional networks for time-series forecasting , author=. International conference on learning representations , year=

work page
[40]

Quarterly of applied Mathematics , volume=

Relaxation oscillations , author=. Quarterly of applied Mathematics , volume=

work page
[41]

Advances in Neural Information Processing Systems , volume=

Sutranets: sub-series autoregressive networks for long-sequence, probabilistic forecasting , author=. Advances in Neural Information Processing Systems , volume=

work page
[42]

International journal of forecasting , volume=

DeepAR: Probabilistic forecasting with autoregressive recurrent networks , author=. International journal of forecasting , volume=. 2020 , publisher=

work page 2020
[43]

Biophysical journal , volume=

Impulses and physiological states in theoretical models of nerve membrane , author=. Biophysical journal , volume=. 1961 , publisher=

work page 1961
[44]

2024 , publisher=

Nonlinear dynamics and chaos: with applications to physics, biology, chemistry, and engineering , author=. 2024 , publisher=

work page 2024
[45]

II , author=

Symmetry breaking instabilities in dissipative systems. II , author=. The Journal of Chemical Physics , volume=. 1968 , publisher=

work page 1968
[47]

Advances in neural information processing systems , volume=

Pytorch: An imperative style, high-performance deep learning library , author=. Advances in neural information processing systems , volume=

work page
[48]

IEEE Open Access Journal of Power and Energy , volume=

Energy forecasting: A review and outlook , author=. IEEE Open Access Journal of Power and Energy , volume=. 2020 , publisher=

work page 2020
[49]

1994 , publisher=

Time series modelling of water resources and environmental systems , author=. 1994 , publisher=

work page 1994
[50]

International conference on learning representations , year=

Reversible instance normalization for accurate time-series forecasting against distribution shift , author=. International conference on learning representations , year=

work page
[51]

Advances in neural information processing systems , volume=

Non-stationary transformers: Exploring the stationarity in time series forecasting , author=. Advances in neural information processing systems , volume=

work page
[52]

Fundamental limitations of foundational forecasting models: The need for multimodality and rigorous evaluation , author=. Proc. NeurIPS Workshop , year=

work page
[54]

Applied soft computing , volume=

Financial time series forecasting with deep learning: A systematic literature review: 2005--2019 , author=. Applied soft computing , volume=. 2020 , publisher=

work page 2005
[55]

Mechanical Systems and Signal Processing , volume=

Statistical pattern recognition for Structural Health Monitoring using time series modeling: Theory and experimental verifications , author=. Mechanical Systems and Signal Processing , volume=. 2009 , publisher=

work page 2009
[56]

European journal of operational research , volume=

Application of machine learning techniques for supply chain demand forecasting , author=. European journal of operational research , volume=. 2008 , publisher=

work page 2008
[57]

IEEE Communications Surveys & Tutorials , volume=

A survey of wearable devices and challenges , author=. IEEE Communications Surveys & Tutorials , volume=. 2017 , publisher=

work page 2017
[58]

Universality in Chaos, 2nd edition , pages=

Deterministic nonperiodic flow 1 , author=. Universality in Chaos, 2nd edition , pages=. 2017 , publisher=

work page 2017
[59]

Breakthroughs in statistics: Methodology and distribution , pages=

Robust estimation of a location parameter , author=. Breakthroughs in statistics: Methodology and distribution , pages=. 1992 , publisher=

work page 1992
[62]

International Conference on Learning Representations , year=

Understanding Straight-Through Estimator in Training Activation Quantized Neural Nets , author=. International Conference on Learning Representations , year=

work page
[63]

Proceedings of the AAAI conference on artificial intelligence , volume=

Informer: Beyond efficient transformer for long sequence time-series forecasting , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

work page
[64]

International Conference on Learning Representations , year=

Categorical Reparameterization with Gumbel-Softmax , author=. International Conference on Learning Representations , year=

work page
[65]

2009 International Joint Conference on Neural Networks , pages=

Long-term prediction of time series by combining direct and mimo strategies , author=. 2009 International Joint Conference on Neural Networks , pages=. 2009 , organization=

work page 2009
[66]

Advances in neural information processing systems , volume=

Film: Frequency improved legendre memory model for long-term time series forecasting , author=. Advances in neural information processing systems , volume=

work page
[67]

The eleventh international conference on learning representations , year=

Micn: Multi-scale local and global context modeling for long-term series forecasting , author=. The eleventh international conference on learning representations , year=

work page
[68]

Proceedings of the AAAI conference on artificial intelligence , volume=

Nhits: Neural hierarchical interpolation for time series forecasting , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

work page
[69]

Long-term Forecasting with Ti

Abhimanyu Das and Weihao Kong and Andrew Leach and Shaan K Mathur and Rajat Sen and Rose Yu , journal=. Long-term Forecasting with Ti. 2023 , note=

work page 2023
[70]

International Journal of forecasting , volume=

The M4 Competition: Results, findings, conclusion and way forward , author=. International Journal of forecasting , volume=. 2018 , publisher=

work page 2018
[71]

IEEE transactions on neural networks and learning systems , volume=

A bias and variance analysis for multistep-ahead time series forecasting , author=. IEEE transactions on neural networks and learning systems , volume=. 2015 , publisher=

work page 2015
[73]

International Conference on Learning Representations , volume=

Fredf: Learning to forecast in the frequency domain , author=. International Conference on Learning Representations , volume=

work page
[74]

Green et al

Stratify: unifying multi-step forecasting strategies: R. Green et al. , author=. Data Mining and Knowledge Discovery , volume=. 2025 , publisher=

work page 2025
[75]

Advances in Neural Information Processing Systems , volume=

Time-o1: Time-series forecasting needs transformed label alignment , author=. Advances in Neural Information Processing Systems , volume=

work page
[76]

Artificial Intelligence Review , volume=

A comprehensive survey of deep learning for time series forecasting: architectural diversity and open challenges , author=. Artificial Intelligence Review , volume=. 2025 , publisher=

work page 2025
[77]

arXiv preprint arXiv:2511.00053 , year=

Quadratic direct forecast for training multi-step time-series forecast models , author=. arXiv preprint arXiv:2511.00053 , year=

work page arXiv
[78]

Lipschitz continuity in model-based reinforcement learning

Kavosh Asadi, Dipendra Misra, and Michael Littman. Lipschitz continuity in model-based reinforcement learning. In International conference on machine learning, pages 264--273. PMLR, 2018

work page 2018
[79]

o ppel, Markus Spanring, Andreas Auer, Oleksandra Prudnikova, Michael Kopp, G \

Maximilian Beck, Korbinian P \"o ppel, Markus Spanring, Andreas Auer, Oleksandra Prudnikova, Michael Kopp, G \"u nter Klambauer, Johannes Brandstetter, and Sepp Hochreiter. xlstm: Extended long short-term memory. Advances in Neural Information Processing Systems, 37: 0 107547--107603, 2024

work page 2024
[80]

Scheduled sampling for sequence prediction with recurrent neural networks

Samy Bengio, Oriol Vinyals, Navdeep Jaitly, and Noam Shazeer. Scheduled sampling for sequence prediction with recurrent neural networks. Advances in neural information processing systems, 28, 2015

work page 2015
[81]

Fundamental limitations of foundational forecasting models: The need for multimodality and rigorous evaluation

Christoph Bergmeir. Fundamental limitations of foundational forecasting models: The need for multimodality and rigorous evaluation. In Proc. NeurIPS Workshop, 2024

work page 2024
[82]

Sutranets: sub-series autoregressive networks for long-sequence, probabilistic forecasting

Shane Bergsma, Tim Zeyl, and Lei Guo. Sutranets: sub-series autoregressive networks for long-sequence, probabilistic forecasting. Advances in Neural Information Processing Systems, 36: 0 30518--30533, 2023

work page 2023
[83]

Lightts: Lightweight time series classification with adaptive ensemble distillation

David Campos, Miao Zhang, Bin Yang, Tung Kieu, Chenjuan Guo, and Christian S Jensen. Lightts: Lightweight time series classification with adaptive ensemble distillation. Proceedings of the ACM on Management of Data, 1: 0 1--27, 2023

work page 2023
[84]

Application of machine learning techniques for supply chain demand forecasting

Real Carbonneau, Kevin Laframboise, and Rustam Vahidov. Application of machine learning techniques for supply chain demand forecasting. European journal of operational research, 184 0 (3): 0 1140--1154, 2008

work page 2008
[85]

Nhits: Neural hierarchical interpolation for time series forecasting

Cristian Challu, Kin G Olivares, Boris N Oreshkin, Federico Garza Ramirez, Max Mergenthaler Canseco, and Artur Dubrawski. Nhits: Neural hierarchical interpolation for time series forecasting. In Proceedings of the AAAI conference on artificial intelligence, volume 37, pages 6989--6997, 2023

work page 2023
[86]

TSM ixer: An all- MLP architecture for time series forecast-ing

Si-An Chen, Chun-Liang Li, Sercan O Arik, Nathanael Christian Yoder, and Tomas Pfister. TSM ixer: An all- MLP architecture for time series forecast-ing. Transactions on Machine Learning Research, 2023. ISSN 2835-8856

work page 2023

Showing first 80 references.