Pith. sign in

REVIEW 4 major objections 6 minor 50 references

ASTER: Adaptive Spatio-Temporal Early Decision Model for Dynamic Resource Allocation

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read ASTER jointly optimizes early spatio-temporal prediction and constrained resource allocation, and reports an average 11.15 percent success-rate gain over the second-best decoupled baseline on four real-world datasets.

desk verdict A well-motivated framework for coupling spatio-temporal prediction with resource allocation, but the SOTA claim is contradicted by the paper's own Table 1 and the evaluation is confounded. read the letter →

arxiv 2506.17929 v1 pith:EYPBQFR6 submitted 2025-06-22 cs.LG cs.AI

Shulun Chen , Wei Shao , Flora D. Salim , Hao Xue This is my paper · ORCID
classification cs.LGcs.AI
keywords spatio-temporalpredictionearlydecisionmakingdynamicresourceallocationmulti-objectivereinforcementlearningresource-awaregraphpredict-then-optimizeemergencyresponsedecision-focused
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 argues that emergency-response and similar dynamic resource-allocation systems lose real-world utility when prediction is treated as a separate stage from allocation: forecasts are made first, then converted into actions, so errors and delays accumulate. To fix this, it proposes ASTER, a model that jointly learns to predict incidents and to allocate limited resources under dynamic availability, so the same representations drive both what will happen and what to do. The central claim, supported by experiments on four real-world datasets, is that this joint approach outperforms decoupled forecast-then-allocate pipelines, with an average 11.15 percent improvement in success rate over the second-best baseline and the best cost-effectiveness ratio on every dataset. A sympathetic reading is that coupling prediction with decision-making, rather than only improving forecast accuracy, is the route to actionable spatio-temporal intelligence.

What carries the argument

Two coupled modules carry the argument. RaST's resource-aware graph learning layer multiplies an asymmetric learned affinity matrix $A^*$ by a binary resource mask $M$, zeroing out edges between busy nodes; its dynamic spatio-temporal convolution layers weight four parallel temporal filters by the resource ratio $\gamma = S_t/S$, so scarce resources shorten the effective temporal horizon, and the state generator fuses long- and short-term encoder outputs with the same $\gamma$ while predicting a node-specific horizon $k = \gamma k_{\max}$. Poda replaces downstream optimization post-processing with a multi-objective Q-network $Q_\theta: \mathcal{S}\times\mathcal{A}\times\Omega\to\mathbb{R}^d$; action selection uses the preference-scalarized value $\omega^\top Q(s,a,\omega)$, and training anneals $\lambda$ from 0 to 0.6 between vector regression and preference-scalarization losses. The same resource state therefore shapes representation, horizon choice, and reward, which is what couples prediction to feasible action.

What would settle it

Re-run the comparison in a simulator where success requires the allocated resource to arrive within a fixed statutory response window rather than merely being assigned before the event, keeping the same data, baselines, resource levels, and evaluation protocol; the central claim fails if ASTER's reported 11.15 percent average success-rate advantage over the second-best approach disappears or reverses.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that early spatio-temporal prediction and downstream resource allocation should be optimized as one coupled system rather than cascaded. ASTER couples two components: RaST, a resource-aware encoder that masks the learned spatial adjacency matrix so message passing only occurs among nodes with idle resources, adapts temporal receptive fields to resource abundance, and fuses long- and short-term encodings by a resource ratio; and Poda, a preference-conditioned multi-objective Q-network $Q_\theta(s,a,\omega)$ that outputs per-objective values and selects actions by $\omega^\top Q$, with hidden preferences inferred from observed trajectories. Under a common evaluation where all baselines feed a fixed rank-and-allocate policy in a resource-constrained simulator, the paper reports an average 11.15 percent improvement in success rate over the second-best approach and the best cost-effectiveness ratio on all four datasets, which it takes as evidence that joint optimization with resource awareness is the effective route to actionable spatio-temporal intelligence.

Load-bearing premise

The paper's own limitation note (Appendix D.1) concedes that real-world constraints are more complex and dynamically evolving than the simulator models, so the load-bearing premise is that the simulated success rule (a resource assigned before an incident counts as a success) and the reward weights in Equation (5) capture true stakeholder trade-offs, and that the fixed rank-and-allocate policy is a fair decoupled baseline; if either is off, the reported 11.15 percent gain measures the simulator's reward, not real-world utility.

Editorial extensions

If this is right

  • Joint training is the active ingredient the paper identifies: coupling the encoder and the decision agent in one objective outperforms feeding the same allocation policy with separately trained forecasters.
  • Resource abundance becomes a planning signal: ASTER extends its horizon when resources are plentiful and contracts to short-term patterns when resources are scarce, a behavior the case studies show preserves resources over time.
  • One preference-conditioned Q-network can cover many stakeholder trade-offs without retraining per preference, and hidden preferences can be recovered from observed behavior when no explicit priority is given.
  • The paper's six-metric evaluation (success rate, false alarm rate, average distance, average early time, resource utilization, cost-effectiveness) offers a common yardstick for early-decision systems that pure forecast error cannot provide.

Reading between the lines

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

  • The paper leaves implicit that the resource-adaptive horizon $k=\gamma k_{\max}$ can be read as an explicit early-decision policy; comparing it with optimal-stopping rules from early time-series classification would test whether the learned horizons match information-theoretic stopping boundaries.
  • Because the training protocol alternates encoder updates and agent updates, one can swap in a different spatio-temporal encoder and retrain Poda on top; if the reported gains transfer across encoders, the joint-coupling mechanism itself—not RaST's specific convolutions—is the active ingredient.
  • The same resource-mask-plus-preference-agent design applies to fleet repositioning, wildfire resource staging, or vaccine allocation; the paper's experiments cover only crime, accidents, EMS dispatch, and traffic incidents.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes ASTER, a model for early spatio-temporal decision-making that jointly learns forecasting and resource allocation under dynamic resource constraints. ASTER combines a Resource-aware Spatio-Temporal interaction module (RaST) with a Preference-oriented decision agent (Poda) trained by multi-objective reinforcement learning. The authors evaluate on four real-world datasets (NYC, NYPD, EMS, XTraffic) with six downstream metrics and claim an average 11.15% improvement in success rate over the second-best baselines, as well as consistent superiority across all datasets and metrics. The manuscript also includes ablation studies, a sensitivity analysis, a case study, and appendices with dataset details, training protocol, and limitation discussion.

Significance. If the central claim were established, ASTER would be a useful contribution: it would demonstrate that jointly optimizing prediction and downstream allocation can outperform decoupled predict-then-allocate pipelines in a resource-constrained setting. The paper introduces a plausible modeling idea—resource-aware graph masking and preference-conditioned decision-making—and provides a code link, pseudo-code, and detailed appendices, which are strengths. However, the empirical evidence as presented does not support the headline claim. The evaluation is confounded because ASTER's training reward shares its components with the reported metrics, the claimed universal superiority is contradicted by the paper's own Table 1, and the reported 11.15% average cannot be reproduced from the table without an explicit aggregation convention. These issues are central to the paper's contribution and require substantial revision.

major comments (4)
  1. [Section 4.2, Table 1] The claim that "ASTER consistently outperforms all baseline methods across diverse datasets and evaluation metrics" is contradicted by Table 1. For example, on XTraffic SR, STAEformer achieves 0.34 versus ASTER's 0.30; on NYC AD, STAEformer achieves 120.65 versus ASTER's 148.35; on NYC AET, ESTGCN achieves 6.76 versus ASTER's 3.26; on NYPD FAR, LSTM achieves 0.02 versus ASTER's 0.25; on EMS FAR, ARIMA achieves 0.04 versus ASTER's 0.16; on XTraffic AD, ESTGCN achieves 26.31 versus ASTER's 43.82; and on XTraffic AET, STEMO achieves 7.08 versus ASTER's 3.35. At least seven cells in the table show that ASTER is not the best. Furthermore, the reported "average 11.15% improvement in success rate over the second-best approaches" is not reproducible from the table. Using the natural definition of second-best as the best non-ASTER SR per dataset, the per-dataset improvements are roughly +28.6%, +20.7%, +12.0%, and (depending on whether the best or second-best baseline is used for XTraffic) either -11.8% or -3.2%, which average to about +12.4% or +14.5%, not 11.15%. The paper must state exactly which cells are considered wins, which aggregation is used, and why "consistently outperforms" is appropriate when multiple cells show otherwise.
  2. [Section 4.1 and Appendix C.2] The evaluation protocol is confounded: all baselines are coupled to a fixed rank-and-allocate policy based on forecasted intensities, while ASTER's Poda is trained to maximize a reward (Eq. 5) that is composed of the same quantities used in the reported metrics (accuracy reward, false alarm reward, distance reward, and temporal reward). This means ASTER is directly optimizing the measurement instrument, whereas the baselines are not. Without an ablation that separates the forecasting component from the allocation component—for example, applying Poda to baseline forecasts, or applying the fixed allocation policy to ASTER's predictions—the reported gains cannot be attributed to the joint forecasting–decision model rather than to the benefit of reward optimization itself. The paper should include such controlled comparisons or explicitly argue why the current comparison is informative despite this confound.
  3. [Section 3.3, Eq. (6)] The multi-objective TD target in Eq. (6) is not well-defined: the vector y is set equal to r_t + gamma * max_{a,omega'} omega^T Q(s_{t+1}, a, omega'; theta), which adds a vector r_t to a scalar. The norm ||y - Q||_2^2 requires y and Q to have the same dimension. The correct vector target would be r_t + gamma * Q(s_{t+1}, a*, omega'; theta) where a* maximizes omega^T Q, or the formulation must be rewritten to make the dimensionality consistent. This is load-bearing because both losses LA and LB depend on y, and a dimension mismatch makes the training objective undefined as written.
  4. [Section 3.3 and Appendix C.4] The hidden-preference inference procedure is nearly tautological for the tested setting: the agent infers omega by maximizing sum_t omega^T r_t, i.e., by selecting the objective that accumulated the largest total reward in a trajectory. When the true preferences are one-hot, any trajectory generated with that preference will tend to have the largest reward on that objective, so the inferred preference will trivially match. The appendix's claim that the agent "successfully infers and selects preferences that closely match the hidden objectives" therefore does not demonstrate genuine preference recovery or adaptation. To make this claim meaningful, the authors should test on non-one-hot preferences, hold out behavioral data for inference, or compare inferred preferences against a ground-truth preference in a setting where the argmax is not the only plausible estimator.
minor comments (6)
  1. [Table 1] The paper reports means for three trials but no standard deviations or significance tests; adding error bars or confidence intervals would help assess whether the differences are meaningful.
  2. [Section 4.1, CER definition] In the Cost-Effectiveness Ratio definition, CER = (1/n) * sum_i r_i / c_i, but the cost c_i is never defined. Please specify how c_i is computed.
  3. [Appendix C.2, baseline details] The text says "All spatio-temporal models are implemented and trained under unified framework, from which we extract their optimal parameters" but does not describe how optimal parameters are selected; please clarify the hyperparameter search procedure.
  4. [Appendix B.2, Algorithm 1] There is a typo: "Observe recoded values" should be "Observe recorded values".
  5. [Table 2] The table header uses "GWNeT" but the model is GraphWaveNet; please correct the capitalization.
  6. [Section 3.2, Eq. (1)] The resource mask M is defined as 1 if "resource in both node i and node j are available," but it is unclear whether this is intended to be symmetric and how it interacts with the asymmetric affinity matrix A*; please clarify.

Circularity Check

2 steps flagged · score 6.0 of 10

Partial circularity: CER metric reuses the training reward and hidden-preference recovery is the argmax of observed return; the Table 1 'consistently outperforms' wording is also internally contradicted, though that is a consistency issue rather than circularity.

  1. self definitional [Section 3.3 (Eq. 5, Poda reward) and Section 4.1 (CER metric definition)]
    "The reward signal rt ={r i t,acc, ri t,false, ri t,distance, ri t,time}n i=1 includes accuracy reward, false alarm reward, allocation distance reward, and temporal reward. The specific calculation is as follows: ri t =α·r i t,acc−β·r i t,false−γ·r i t,distance +δ·r i t,time. ... Cost-Effectiveness Ratio(CER): Quantifies average reward per unit resource allocation cost, CER= 1 n Pn i=1 ri ci , whereri andci denote reward and allocation cost."

    The only reward defined in the paper is the Eq. (5) scalarized reward that Poda is explicitly trained to maximize through the losses in Eq. (6). The CER metric is then defined using the same symbol r_i divided by cost. Thus CER is not an independent downstream-utility measure: it is the training objective itself, normalized by allocation cost. Reporting ASTER's 'comprehensive superiority in downstream utility metrics such as RUR and CER' is therefore partly forced for CER, because ASTER optimizes exactly the quantity being scored. The other five metrics retain independent content, so this is a partial, not total, circularity.

  2. fitted input called prediction [Section 3.3 ('Inference of Hidden Preferences') and Appendix C.4]
    "For each trajectory, we optimizeω to maximize the expected scalarized return:ω∗ = arg maxω∈Ω P tω⊤rt. ... As shown in Table 4, the agent successfully infers and selects preferences that closely match the hidden objectives, indicating effective alignment with the true underlying priorities."

    The inferred preference ω* is defined as the maximizer of Σ_t ω^T r_t over the preference simplex. For any fixed trajectory, this argmax is always the one-hot vector of the objective with the largest cumulative reward. The Appendix C.4 experiment plants one-hot hidden preferences and then reports that the inferred preference matches them, but this is the estimator's own definition: the 'recovered' preference is just the max-reward component. It is not evidence that an independent latent user preference was discovered from behavior. If the planted preference were not the component with the largest cumulative reward, ω* would not match it, so the experiment only verifies a tautology rather than a learned inference capability.

full rationale

The core ASTER derivation—RaST encoding, the Poda multi-objective RL agent, and the comparison against nine forecasting/early-prediction baselines under a shared allocation policy—is not tautological: the architecture and the SR, FAR, AD, AET, and RUR metrics have independent content. However, two load-bearing evaluation steps reduce to the paper's own definitions. First, CER is defined with r_i, the same scalarized reward that Poda is trained to maximize (Eq. 5), so the CER advantage is a relabeled training objective rather than an independent utility result. Second, hidden-preference inference defines ω* as argmax of cumulative scalarized return, making the Appendix C.4 'successful inference' of one-hot preferences a direct consequence of the estimator's definition. These are concrete reductions by construction, not vague impressions. The additional Table 1 inconsistency—ASTER is not best in multiple metric–dataset cells and the 11.15% average is not reproducible from the shown numbers—is a correctness/consistency problem rather than a circularity pattern, so I do not count it as a separate circular step, but it further weakens the paper's headline claim. No load-bearing self-citation chain or imported uniqueness theorem was found; the cited STEMO/ESTGCN works by co-authors are used as baselines and related work, not as proof of ASTER's superiority. Score 6 reflects partial circularity: some 'predictions' reduce by construction, while the central framework still has independent empirical content, so it is not 8–10.

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

The reported gains rest on a hand-built simulator, reward weights that are never disclosed, and hyperparameters tuned on one dataset. The model introduces no new physical entities, but its evaluation is self-referential because the training reward and the success metric share the same definitions.

free parameters (4)
  • reward coefficients alpha, beta, gamma, delta = not reported
    Eq. (5) weights the four reward terms; the paper never states their values, yet they determine the trained policy and enter CER through r_i.
  • maximum prediction horizon k_max = 12
    Selected in Appendix C.3 by sensitivity analysis on EMS; k = round((S_t/S)*k_max) directly controls the masked prediction loss and decision horizon.
  • number of attention heads h = 4
    Selected in Appendix C.3 by sensitivity analysis on EMS; used in the state generator's MHA.
  • resource level thresholds = per-dataset S ranges (e.g., NYC low S<=53, medium <=88)
    Hand-defined in Appendix C.1 to create low/medium/high scenarios; the robustness conclusion depends on these arbitrary partitions.
assumptions (4)
  • domain assumption An allocated resource at node i before an incident counts as a successful intervention; success requires both the prediction and the action.
    This proxy defines SR/FAR and the accuracy reward (Section 3.3, Section 4.1). It ignores whether the resource could actually prevent the incident.
  • domain assumption The discrete-time environment with Hungarian matching, cooldowns, and a distance matrix faithfully simulates real emergency dispatching.
    Used in Section 3.3 and Appendix C.2 to generate transitions and compute AD and CER; cooldown lengths and demand models are not specified.
  • ad hoc to paper Message passing should only occur between nodes with simultaneously available resources.
    Eq. (1) applies the mask M to the learned adjacency; this filters out all edges incident to busy nodes, which is a design choice, not a derived result.
  • ad hoc to paper The hidden preference can be recovered by maximizing the sum of scalarized returns over a trajectory.
    Section 3.3 defines omega* = argmax sum omega^T r_t; this by construction selects the objective with the largest observed return, so Appendix C.4's 'recovery' of one-hot preferences is tautological.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ASTER: Adaptive Spatio-Temporal Early Decision Model for Dynamic Resource Allocation." pith.science (2026). https://pith.science/paper/EYPBQFR6

@misc{pith2026250617929,
  author       = {Pith},
  title        = {Pith review of: ASTER: Adaptive Spatio-Temporal Early Decision Model for Dynamic Resource Allocation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EYPBQFR6}},
  note         = {Machine review of arXiv:2506.17929}
}
read the original abstract

Supporting decision-making has long been a central vision in the field of spatio-temporal intelligence. While prior work has improved the timeliness and accuracy of spatio-temporal forecasting, converting these forecasts into actionable strategies remains a key challenge. A main limitation is the decoupling of the prediction and the downstream decision phases, which can significantly degrade the downstream efficiency. For example, in emergency response, the priority is successful resource allocation and intervention, not just incident prediction. To this end, it is essential to propose an Adaptive Spatio-Temporal Early Decision model (ASTER) that reforms the forecasting paradigm from event anticipation to actionable decision support. This framework ensures that information is directly used for decision-making, thereby maximizing overall effectiveness. Specifically, ASTER introduces a new Resource-aware Spatio-Temporal interaction module (RaST) that adaptively captures long- and short-term dependencies under dynamic resource conditions, producing context-aware spatiotemporal representations. To directly generate actionable decisions, we further design a Preference-oriented decision agent (Poda) based on multi-objective reinforcement learning, which transforms predictive signals into resource-efficient intervention strategies by deriving optimal actions under specific preferences and dynamic constraints. Experimental results on four benchmark datasets demonstrate the state-of-the-art performance of ASTER in improving both early prediction accuracy and resource allocation outcomes across six downstream metrics.

Figures

Figures reproduced from arXiv: 2506.17929 by the authors.

Figure 1
Figure 1. Example of three methods for early spatio-temporal decision-making. Each circle is a node (e.g., regions) with a recorded signal (e.g., ac￾cident occurrence) over time. In the result column, green nodes indicate successful interventions, red nodes represent incidents without intervention, and yellow nodes correspond to false alarms. While some of these strategies offer unique ad￾vantages, they still face critical li… view at source ↗
Figure 2
Figure 2. Framework overview of ASTER. the historical sequence X = [Xt−L, . . . , Xt−1, Xt] ∈ R L×N×C , a spatio-temporal model takes as input features X and produces outputs Xˆ T . These outputs are then used to parameterize an optimization problem that is solved to yield actionable decisions. Specifically, spatio-temporal decision-making aims to select a binary action vector a = {ai} n i=1 ∈ {0, 1} n, where ai = 1 indicates… view at source ↗
Figure 3
Figure 3. Ablation studies of ASTER on NYPD, AD and AET are linearly rescaled. The results of "w/o dual encoder" indicate that capturing resource information along with the corresponding spatio-temporal dependen￾cies is essential for supporting effective down￾stream actions. The performance of "w/o state generator" further highlights the effectiveness of dynamically fusing long- and short-term representations based on real-ti… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Case studies of ASTER on NYPD and EMS dataset. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: This figure illustrates the process of Spatio-Temporal Representation and Integration within [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: We report how changes in hyperparameters affect SR (Success Rate) and CER (Cost-Effectiveness Ratio). SR reflects the model’s ability to make successful in￾terventions, while CER captures the overall efficiency of downstream resource utilization. 1 2 4 8 16 0.2 0.3 0.4…

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 33 canonical work pages

  1. [1]

    Denis, R Chelsea Nagy, Virginia Iglesias, John T Abatzoglou, and Jennifer K Balch

    Maxwell B Joseph, Matthew W Rossi, Nathan P Mietkiewicz, Adam L Mahood, Megan E Cattau, Lise Ann St. Denis, R Chelsea Nagy, Virginia Iglesias, John T Abatzoglou, and Jennifer K Balch. Spatiotemporal prediction of wildfire size extremes with bayesian finite sample maxima. Ecological Applications, 29(6):e01898, 2019

    work page 2019

  2. [2]

    Long-term spatio-temporal forecasting via dynamic multiple-graph attention.arXiv preprint arXiv:2204.11008, 2022

    Wei Shao, Zhiling Jin, Shuo Wang, Yufan Kang, Xiao Xiao, Hamid Menouar, Zhaofeng Zhang, Junshan Zhang, and Flora Salim. Long-term spatio-temporal forecasting via dynamic multiple-graph attention.arXiv preprint arXiv:2204.11008, 2022

    arXiv 2022

  3. [3]

    Transferrable contextual feature clusters for parking occupancy prediction.Pervasive and Mobile Computing, 97:101831, 2024

    Wei Shao, Yu Zhang, Pengfei Xiao, Kyle Kai Qin, Mohammad Saiedur Rahaman, Jeffrey Chan, Bin Guo, Andy Song, and Flora D Salim. Transferrable contextual feature clusters for parking occupancy prediction.Pervasive and Mobile Computing, 97:101831, 2024

    work page 2024

  4. [4]

    Spatial-Temporal Sequential Hypergraph Network for Crime Prediction with Dynamic Multiplex Relation Learning

    Lianghao Xia, Chao Huang, Yong Xu, Peng Dai, Liefeng Bo, Xiyue Zhang, and Tianyi Chen. Spatial-temporal sequential hypergraph network for crime prediction with dynamic multiplex relation learning.arXiv preprint arXiv:2201.02435, 2022

    work page Pith review arXiv 2022

  5. [5]

    Recurrent halting chain for early multi-label classification

    Thomas Hartvigsen, Cansu Sen, Xiangnan Kong, and Elke Rundensteiner. Recurrent halting chain for early multi-label classification. InProceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 1382–1392, 2020

    work page 2020

  6. [6]

    Diffusion convolutional recurrent neural network: Data-driven traffic forecasting.arXiv preprint arXiv:1707.01926, 2017

    Yaguang Li, Rose Yu, Cyrus Shahabi, and Yan Liu. Diffusion convolutional recurrent neural network: Data-driven traffic forecasting.arXiv preprint arXiv:1707.01926, 2017

    arXiv 2017

  7. [7]

    Early spatiotemporal event prediction via adaptive controller and spatiotemporal embedding

    Wei Shao, Ziyan Peng, Yufan Kang, Xiao Xiao, and Zhiling Jin. Early spatiotemporal event prediction via adaptive controller and spatiotemporal embedding. In2023 IEEE International Conference on Data Mining (ICDM), pages 1307–1312. IEEE, 2023

    work page 2023

  8. [8]

    Stemo: Early spatio-temporal forecasting with multi-objective reinforcement learning

    Wei Shao, Yufan Kang, Ziyan Peng, Xiao Xiao, Lei Wang, Yuhui Yang, and Flora D Salim. Stemo: Early spatio-temporal forecasting with multi-objective reinforcement learning. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 2618–2627, 2024

    work page 2024

Show all 50 references
  1. [9]

    predict, then optimize

    Adam N Elmachtoub and Paul Grigas. Smart “predict, then optimize”.Management Science, 68(1):9–26, 2022

    2022

  2. [10]

    Decision-focused learning without decision-making: Learning locally optimized decision losses.Advances in Neural Information Processing Systems, 35:1320–1332, 2022

    Sanket Shah, Kai Wang, Bryan Wilder, Andrew Perrault, and Milind Tambe. Decision-focused learning without decision-making: Learning locally optimized decision losses.Advances in Neural Information Processing Systems, 35:1320–1332, 2022

    2022

  3. [11]

    Traveling officer problem: Managing car parking violations efficiently using sensor data.IEEE Internet of Things Journal, 5(2):802–810, 2017

    Wei Shao, Flora D Salim, Tao Gu, Ngoc-Thanh Dinh, and Jeffrey Chan. Traveling officer problem: Managing car parking violations efficiently using sensor data.IEEE Internet of Things Journal, 5(2):802–810, 2017

    2017

  4. [12]

    Incorporating lstm auto-encoders in optimizations to solve parking officer patrolling problem

    Wei Shao, Siyu Tan, Sichen Zhao, Kyle Kai Qin, Xinhong Hei, Jeffrey Chan, and Flora D Salim. Incorporating lstm auto-encoders in optimizations to solve parking officer patrolling problem. ACM Transactions on Spatial Algorithms and Systems (TSAS), 6(3):1–21, 2020

    2020

  5. [13]

    Sepp hochreiter and jürgen schmidhuber.Neural Computation, 9(8):1735, 1997

    Long Short-Term Memory. Sepp hochreiter and jürgen schmidhuber.Neural Computation, 9(8):1735, 1997

    1997

  6. [14]

    Quantifying uncertainty in deep spatiotemporal forecasting

    Dongxia Wu, Liyao Gao, Matteo Chinazzi, Xinyue Xiong, Alessandro Vespignani, Yi-An Ma, and Rose Yu. Quantifying uncertainty in deep spatiotemporal forecasting. InProceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining, pages 1841–1851, 2021

    2021

  7. [15]

    Graph wavenet for deep spatial-temporal graph modeling.arXiv preprint arXiv:1906.00121, 2019

    Zonghan Wu, Shirui Pan, Guodong Long, Jing Jiang, and Chengqi Zhang. Graph wavenet for deep spatial-temporal graph modeling.arXiv preprint arXiv:1906.00121, 2019

    1906 arXiv

  8. [16]

    Mtgnn: multi-task graph neural network based few-shot learning for disease similarity measurement.Methods, 198:88–95, 2022

    Jianliang Gao, Xiangchi Zhang, Ling Tian, Yuxin Liu, Jianxin Wang, Zhao Li, and Xiaohua Hu. Mtgnn: multi-task graph neural network based few-shot learning for disease similarity measurement.Methods, 198:88–95, 2022. 10

    2022

  9. [17]

    Gman: A graph multi- attention network for traffic prediction

    Chuanpan Zheng, Xiaoliang Fan, Cheng Wang, and Jianzhong Qi. Gman: A graph multi- attention network for traffic prediction. InProceedings of the AAAI conference on artificial intelligence, volume 34, pages 1234–1241, 2020

    2020

  10. [18]

    Attention based spatial- temporal graph convolutional networks for traffic flow forecasting

    Shengnan Guo, Youfang Lin, Ning Feng, Chao Song, and Huaiyu Wan. Attention based spatial- temporal graph convolutional networks for traffic flow forecasting. InProceedings of the AAAI conference on artificial intelligence, volume 33, pages 922–929, 2019

    2019

  11. [19]

    Adaptive graph convolutional recurrent network for traffic forecasting.Advances in neural information processing systems, 33:17804–17815, 2020

    Lei Bai, Lina Yao, Can Li, Xianzhi Wang, and Can Wang. Adaptive graph convolutional recurrent network for traffic forecasting.Advances in neural information processing systems, 33:17804–17815, 2020

    2020

  12. [20]

    Spatio-temporal pivotal graph neural networks for traffic flow forecasting

    Weiyang Kong, Ziyu Guo, and Yubao Liu. Spatio-temporal pivotal graph neural networks for traffic flow forecasting. InProceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 8627–8635, 2024

    2024

  13. [21]

    Pdformer: Propagation delay-aware dynamic long-range transformer for traffic flow prediction

    Jiawei Jiang, Chengkai Han, Wayne Xin Zhao, and Jingyuan Wang. Pdformer: Propagation delay-aware dynamic long-range transformer for traffic flow prediction. InProceedings of the AAAI conference on artificial intelligence, volume 37, pages 4365–4373, 2023

    2023

  14. [22]

    Spatial-temporal synchronous graph convolutional networks: A new framework for spatial-temporal network data forecasting

    Chao Song, Youfang Lin, Shengnan Guo, and Huaiyu Wan. Spatial-temporal synchronous graph convolutional networks: A new framework for spatial-temporal network data forecasting. InProceedings of the AAAI conference on artificial intelligence, volume 34, pages 914–921, 2020

    2020

  15. [23]

    Stg2seq: Spatial- temporal graph to sequence model for multi-step passenger demand forecasting.arXiv preprint arXiv:1905.10069, 2019

    Lei Bai, Lina Yao, Salil Kanhere, Xianzhi Wang, Quan Sheng, et al. Stg2seq: Spatial- temporal graph to sequence model for multi-step passenger demand forecasting.arXiv preprint arXiv:1905.10069, 2019

    1905 arXiv

  16. [24]

    Pre-training enhanced spatial-temporal graph neural network for multivariate time series forecasting

    Zezhi Shao, Zhao Zhang, Fei Wang, and Yongjun Xu. Pre-training enhanced spatial-temporal graph neural network for multivariate time series forecasting. InProceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining, pages 1567–1577, 2022

    2022

  17. [25]

    Gpt-st: generative pre-training of spatio-temporal graph neural networks.Advances in neural information processing systems, 36:70229–70246, 2023

    Zhonghang Li, Lianghao Xia, Yong Xu, and Chao Huang. Gpt-st: generative pre-training of spatio-temporal graph neural networks.Advances in neural information processing systems, 36:70229–70246, 2023

    2023

  18. [26]

    Utilizing temporal patterns for estimating uncertainty in interpretable early decision making

    Mohamed F Ghalwash, Vladan Radosavljevic, and Zoran Obradovic. Utilizing temporal patterns for estimating uncertainty in interpretable early decision making. InProceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining, pages 402–411, 2014

    2014

  19. [27]

    Early classification on multivariate time series.Neurocomputing, 149:777–787, 2015

    Guoliang He, Yong Duan, Rong Peng, Xiaoyuan Jing, Tieyun Qian, and Lingling Wang. Early classification on multivariate time series.Neurocomputing, 149:777–787, 2015

    2015

  20. [28]

    Early classification of time series by simultaneously optimizing the accuracy and earliness.IEEE transactions on neural networks and learning systems, 29(10):4569–4578, 2017

    Usue Mori, Alexander Mendiburu, Sanjoy Dasgupta, and Jose A Lozano. Early classification of time series by simultaneously optimizing the accuracy and earliness.IEEE transactions on neural networks and learning systems, 29(10):4569–4578, 2017

    2017

  21. [29]

    Reliable early classifi- cation of time series based on discriminating the classes over time.Data mining and knowledge discovery, 31:233–263, 2017

    Usue Mori, Alexander Mendiburu, Eamonn Keogh, and Jose A Lozano. Reliable early classifi- cation of time series based on discriminating the classes over time.Data mining and knowledge discovery, 31:233–263, 2017

    2017

  22. [30]

    Early classification of multi- variate time series using a hybrid hmm/svm model

    Mohamed F Ghalwash, Dušan Ramljak, and Zoran Obradovi´c. Early classification of multi- variate time series using a hybrid hmm/svm model. In2012 IEEE International Conference on Bioinformatics and Biomedicine, pages 1–6. IEEE, 2012

    2012

  23. [31]

    Spatio-temporal adaptive embedding makes vanilla transformer sota for traffic forecasting

    Hangchen Liu, Zheng Dong, Renhe Jiang, Jiewen Deng, Jinliang Deng, Quanjun Chen, and Xuan Song. Spatio-temporal adaptive embedding makes vanilla transformer sota for traffic forecasting. InProceedings of the 32nd ACM international conference on information and knowledge manage...

    2023

  24. [32]

    Cost-aware early classification of time series

    Romain Tavenard and Simon Malinowski. Cost-aware early classification of time series. In Machine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2016, Riva del Garda, Italy, September 19-23, 2016, Proceedings, Part I 16, pages 632–647. Springer, 2016

    2016

  25. [33]

    End to end learning and optimization on graphs.Advances in Neural Information Processing Systems, 32, 2019

    Bryan Wilder, Eric Ewing, Bistra Dilkina, and Milind Tambe. End to end learning and optimization on graphs.Advances in Neural Information Processing Systems, 32, 2019

    2019

  26. [34]

    Supply- demand-aware deep reinforcement learning for dynamic fleet management.ACM Transactions on Intelligent Systems and Technology (TIST), 13(3):1–19, 2022

    Bolong Zheng, Lingfeng Ming, Qi Hu, Zhipeng Lü, Guanfeng Liu, and Xiaofang Zhou. Supply- demand-aware deep reinforcement learning for dynamic fleet management.ACM Transactions on Intelligent Systems and Technology (TIST), 13(3):1–19, 2022

    2022

  27. [35]

    Difflight: a partial rewards conditioned diffusion model for traffic signal control with missing data.Advances in Neural Information Processing Systems, 37:123353–123378, 2024

    Hanyang Chen, Yang Jiang, Shengnan Guo, Xiaowei Mao, Youfang Lin, and Huaiyu Wan. Difflight: a partial rewards conditioned diffusion model for traffic signal control with missing data.Advances in Neural Information Processing Systems, 37:123353–123378, 2024

    2024

  28. [36]

    Reinforcement learning enhances the experts: Large-scale covid-19 vaccine allocation with multi-factor contact network

    Qianyue Hao, Wenzhen Huang, Fengli Xu, Kun Tang, and Yong Li. Reinforcement learning enhances the experts: Large-scale covid-19 vaccine allocation with multi-factor contact network. InProceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 4...

    2022

  29. [37]

    Pointer networks.Advances in neural information processing systems, 28, 2015

    Oriol Vinyals, Meire Fortunato, and Navdeep Jaitly. Pointer networks.Advances in neural information processing systems, 28, 2015

    2015

  30. [38]

    A generalized algorithm for multi- objective reinforcement learning and policy adaptation.Advances in neural information pro- cessing systems, 32, 2019

    Runzhe Yang, Xingyuan Sun, and Karthik Narasimhan. A generalized algorithm for multi- objective reinforcement learning and policy adaptation.Advances in neural information pro- cessing systems, 32, 2019

    2019

  31. [39]

    Pareto set learning for expensive multi-objective optimization.Advances in neural information processing systems, 35:19231– 19247, 2022

    Xi Lin, Zhiyuan Yang, Xiaoyuan Zhang, and Qingfu Zhang. Pareto set learning for expensive multi-objective optimization.Advances in neural information processing systems, 35:19231– 19247, 2022

    2022

  32. [40]

    Rewarded soups: towards pareto-optimal alignment by interpolating weights fine-tuned on diverse rewards.Advances in Neural Information Processing Systems, 36:71095–71134, 2023

    Alexandre Rame, Guillaume Couairon, Corentin Dancette, Jean-Baptiste Gaya, Mustafa Shukor, Laure Soulier, and Matthieu Cord. Rewarded soups: towards pareto-optimal alignment by interpolating weights fine-tuned on diverse rewards.Advances in Neural Information Processing System...

    2023

  33. [41]

    Multi-scale context aggregation by dilated convolutions.arXiv preprint arXiv:1511.07122, 2015

    Fisher Yu and Vladlen Koltun. Multi-scale context aggregation by dilated convolutions.arXiv preprint arXiv:1511.07122, 2015

    2015 arXiv

  34. [42]

    End-to-end constrained optimization learning: A survey.arXiv preprint arXiv:2103.16378, 2021

    James Kotary, Ferdinando Fioretto, Pascal Van Hentenryck, and Bryan Wilder. End-to-end constrained optimization learning: A survey.arXiv preprint arXiv:2103.16378, 2021

    2021 arXiv

  35. [43]

    https://www.kaggle.com/datasets/mysarahmadbhat/nyc-traffic-accidents

    NYC. https://www.kaggle.com/datasets/mysarahmadbhat/nyc-traffic-accidents

  36. [44]

    https://data.cityofnewyork.us/Public-Safety/NYPD-Complaint-Data-Historic/qgea-i56i

    NYPD. https://data.cityofnewyork.us/Public-Safety/NYPD-Complaint-Data-Historic/qgea-i56i

  37. [45]

    https://data.cityofnewyork.us/Public-Safety/EMS-Incident-Dispatch-Data/76xm-jjuj

    EMS. https://data.cityofnewyork.us/Public-Safety/EMS-Incident-Dispatch-Data/76xm-jjuj

  38. [46]

    Xtraffic: A dataset where traffic meets incidents with explainability and more.arXiv preprint arXiv:2407.11477, 2024

    Xiaochuan Gou, Ziyue Li, Tian Lan, Junpeng Lin, Zhishuai Li, Bingyu Zhao, Chen Zhang, Di Wang, and Xiangliang Zhang. Xtraffic: A dataset where traffic meets incidents with explainability and more.arXiv preprint arXiv:2407.11477, 2024

    2024

  39. [47]

    A summary of traffic flow forecasting methods.Journal of highway and transportation research and development, 21(3):82–85, 2004

    Jing Liu and Wei Guan. A summary of traffic flow forecasting methods.Journal of highway and transportation research and development, 21(3):82–85, 2004

    2004

  40. [48]

    John Wiley & Sons, 2015

    George EP Box, Gwilym M Jenkins, Gregory C Reinsel, and Greta M Ljung.Time series analysis: forecasting and control. John Wiley & Sons, 2015

    2015

  41. [49]

    Adaptive-halting policy network for early classification

    Thomas Hartvigsen, Cansu Sen, Xiangnan Kong, and Elke Rundensteiner. Adaptive-halting policy network for early classification. InProceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 101–110, 2019. 12 Supplementary Material for...

    2019

  42. [50]

    However, excessive extension of this horizon amplifies temporal noise and degrades performance across key metrics, particularly the Cost-Effectiveness Ratio. These results suggest that an appropriately chosen prediction horizon not only improves the model’s ability to balance ...

Pith tools

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