Pith. sign in

REVIEW 2 major objections 1 minor 55 references

A hypernetwork meta learner lets ETA models adapt to entirely new delivery regions without historical data or consistent features.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.3

2026-06-28 17:44 UTC pith:QLAILOS6

load-bearing objection UME proposes a hypernetwork meta-learner in a dual-branch setup to unify cross-domain ETA modeling and handle missing offline features in unseen domains, but the abstract supplies zero quantitative results or technical details to support the claims. the 2 major comments →

arxiv 2606.00979 v1 pith:QLAILOS6 submitted 2026-05-31 cs.LG

UME: A Unified Meta-Generalization Framework for Cross-Domain ETA

Duo Wang , Qiong Wu , Jianguo Wu , Ruiyu Xu , Jinhui Yi , Zhonggen Sun , Zhentao Zhang , Yu Zhang
show 4 more authors
Ke Xing Yongjun Yin Zishuo Li Jianwen Huang
This is my paper
classification cs.LG
keywords meta-learningcross-domain generalizationETA predictionhypernetworkcold-start adaptationmulti-domain modelingknowledge distillationdelivery logistics
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper sets out to create one model that can predict delivery arrival times across many regions and also work immediately in brand-new regions that have never been seen before. Current multi-domain methods either cannot handle completely new areas or force companies to run separate systems when new regions lack the usual statistical features, which raises costs and blocks knowledge sharing. UME combines a dual-branch network with a hypernetwork that reads domain-level patterns and current instance details to adjust feature selection, expert weighting, and the final output on the fly. This design supports zero-shot use in cold-start domains and includes knowledge distillation to boost accuracy. The authors report that the system has been deployed on a major international delivery platform and beats prior baselines in both offline tests and live experiments.

Core claim

UME integrates a unified dual-branch architecture with a hypernetwork-based meta learner. By leveraging domain-level knowledge and instance-level context, the meta learner dynamically modulates feature gating, expert attention, and final prediction to capture cross-domain correlations and enable intra-domain adaptation for unseen domains, even when offline features are structurally missing. A knowledge distillation strategy is added to further improve performance.

What carries the argument

The hypernetwork-based meta learner that uses domain-level knowledge and instance-level context to dynamically modulate feature gating, expert attention, and final prediction.

Load-bearing premise

A hypernetwork can learn to capture cross-domain correlations and adapt within a new domain using only domain-level knowledge plus current instance context, even when key historical statistical features are absent.

What would settle it

Introduce a new test domain whose feature statistics differ sharply from all training domains, remove the usual offline features, and measure whether UME's accuracy falls below a simple baseline trained only on the available data for that domain.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Zero-shot prediction becomes possible for completely unseen domains during the initial cold-start phase.
  • Structural missingness of offline features in new domains no longer requires separate modeling pipelines.
  • Knowledge transfer occurs between mature and cold-start domains inside a single maintained system.
  • Knowledge distillation can be applied within the same meta-generalization loop to raise overall accuracy.
  • A single deployed model can serve international platforms with diverse regional patterns.

Where Pith is reading between the lines

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

  • The same modulation approach could be tested on other prediction tasks that face sudden domain shifts, such as demand forecasting across cities.
  • Eliminating separate domain models might cut long-term engineering overhead in any large-scale logistics or recommendation system.
  • An experiment that adds synthetic domains with extreme feature divergence would show the practical limits of relying on domain-level knowledge alone.
  • Combining this hypernetwork with other forms of context encoding might improve robustness when instance-level signals are also sparse.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 1 minor

Summary. The manuscript proposes UME, a unified meta-generalization framework for cross-domain ETA prediction in instant logistics. It integrates a dual-branch architecture with a hypernetwork-based meta learner that uses domain-level knowledge and instance-level context to dynamically modulate feature gating, expert attention, and final prediction, addressing generalization to unseen domains, structural missingness of offline features, and separate modeling of mature versus cold-start domains. A knowledge distillation strategy is included. The framework is reported as deployed in the Meituan-keeta platform, with claims of significant outperformance over baselines in offline experiments and online A/B tests.

Significance. If the empirical claims and the meta-learner's ability to compensate for missing features hold, the work would have practical significance for industrial multi-domain prediction systems in logistics, potentially reducing maintenance costs by enabling a single model across domains with varying data maturity.

major comments (2)
  1. [Abstract] Abstract: the central claim that the hypernetwork-based meta learner enables zero-shot prediction for completely unseen domains with structurally missing offline features rests on an unverified assumption about cross-domain correlation capture using only domain-level knowledge and instance-level context; no details on hypernetwork input construction, training objective, or experiments with forced feature removal in held-out domains are supplied, which is load-bearing for the unified modeling claim over separate mature/cold-start systems.
  2. [Abstract] Abstract: the assertion that UME significantly outperforms existing baselines in offline experiments and online A/B tests supplies no quantitative results, dataset descriptions, ablation studies, or derivation details, so there is no visible evidence that the described architecture supports the performance claims.
minor comments (1)
  1. [Abstract] Abstract: 'a knowledge distillation strategy is further introduce' contains a grammatical error ('introduce' should be 'introduced').

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful review and constructive feedback. We address the major comments point by point below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the hypernetwork-based meta learner enables zero-shot prediction for completely unseen domains with structurally missing offline features rests on an unverified assumption about cross-domain correlation capture using only domain-level knowledge and instance-level context; no details on hypernetwork input construction, training objective, or experiments with forced feature removal in held-out domains are supplied, which is load-bearing for the unified modeling claim over separate mature/cold-start systems.

    Authors: We agree that the abstract does not supply these technical details. The manuscript body describes the hypernetwork input construction from domain-level knowledge and instance-level context, the training objective, and the experiments with forced feature removal in held-out domains that support the zero-shot claim. We will revise the abstract to briefly reference these elements and strengthen the presentation of the unified modeling approach. revision: yes

  2. Referee: [Abstract] Abstract: the assertion that UME significantly outperforms existing baselines in offline experiments and online A/B tests supplies no quantitative results, dataset descriptions, ablation studies, or derivation details, so there is no visible evidence that the described architecture supports the performance claims.

    Authors: The abstract is written at a high level due to length constraints and therefore omits specific quantitative results and dataset details. The manuscript contains the requested quantitative results, dataset descriptions, ablation studies, and derivation details in the experimental sections. We will revise the abstract to incorporate key quantitative highlights to make the performance claims more directly supported within the abstract itself. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained with external empirical claims

full rationale

The abstract and available description introduce UME as a proposed architecture combining dual-branch design, hypernetwork meta-learner, and knowledge distillation, but contain no equations, training objectives, parameter-fitting steps, or self-citations. Claims of outperforming baselines rest on offline experiments, online A/B tests, and platform deployment, which are presented as externally falsifiable results rather than reductions to inputs by construction. No self-definitional loops, fitted inputs renamed as predictions, or uniqueness theorems imported from prior author work appear in the text. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities; the meta learner and hypernetwork are described at a conceptual level without implementation details.

pith-pipeline@v0.9.1-grok · 5861 in / 1153 out tokens · 24861 ms · 2026-06-28T17:44:39.507336+00:00 · methodology

0 comments
read the original abstract

Accurate Estimated Time of Arrival (ETA) prediction on checkout page is crucial in instant logistics for enhancing user satisfaction, optimizing dispatching, and controlling operational costs. In international on-demand delivery platforms, where ETA data originates from diverse countries or regions with different patterns, multi-domain modeling is of great importance and has been widely adopted. However, existing methods still face three critical challenges in real-world deployment. First, current multi-domain models struggle to generalize to completely unseen domains, failing to achieve zero-shot prediction during the initial cold-start phase. Second, cross-domain feature spaces are often assumed to be consistent, whereas new domains commonly suffer from structural missingness of offline (statistical) features due to the lack of historical data. Third, such feature missingness often compels industrial systems to model mature and cold-start domains separately, hindering knowledge transfer and increasing maintenance overhead. To address these challenges, we propose \textbf{UME}, a \textbf{U}nified \textbf{M}eta-generalization framework for \textbf{E}TA. Specifically, UME integrates a unified dual-branch architecture with a novel meta-learning mechanism that employs a hypernetwork-based meta learner. By leveraging domain-level knowledge and instance-level context, the meta learner empowers three meta modules to dynamically modulate feature gating, expert attention, and final prediction, capturing cross-domain correlations and facilitating intra-domain adaptation. A knowledge distillation strategy is further introduce to enhance performance. UME has now been deployed in Meituan-keeta delivery platform (the largest international food delivery platform in China). Extensive offline experiments and online A/B tests demonstrate that UME significantly outperforms existing baselines.

Figures

Figures reproduced from arXiv: 2606.00979 by Duo Wang, Jianguo Wu, Jianwen Huang, Jinhui Yi, Ke Xing, Qiong Wu, Ruiyu Xu, Yongjun Yin, Yu Zhang, Zhentao Zhang, Zhonggen Sun, Zishuo Li.

Figure 1
Figure 1. Figure 1: Checkout Page ETA However, existing MDL approaches still face three key limita￾tions. First, most MDL methods struggle to generalize to completely unseen domains [31]. As illustrated in [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Challenges in cross-domain ETA [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of UME, illustrating the Unified Dual-Branch Network empowered by a novel meta-learning mechanism. 4.1.1 Unified Dual-Branch Network. In this section, we introduce UDBN, which serve as the feature extraction backbone of our frame￾work. Although the two branches share the same backbone archi￾tecture, they operate under an information asymmetric setting. Feature Encoding and Embedding. As illustrate… view at source ↗
Figure 4
Figure 4. Figure 4: Unified Backbone with Meta Gating 4.1.2 Meta learning Mechanism. To achieve multi-domain model￾ing and cross-domain generalization, we introduce a noval meta￾learning mechanism in which a hypernetwork-based meta learner is designed to learn explicit inter-domain correlations and empowers three meta modules to adaptively modulate UDBN in a domain￾specific manner. Distinct from conventional parameter-generat… view at source ↗
Figure 5
Figure 5. Figure 5: Meta Module adjusting the importance of the 𝐾 (𝐾 = 3) shared experts and one branch-specific expert. Let H𝑠ℎ𝑎𝑟𝑒𝑑 = {ℎ (1) 𝑠ℎ , ℎ(2) 𝑠ℎ , . . . , ℎ(𝐾) 𝑠ℎ } denote the set of outputs from the 𝐾 shared experts. For each branch, we design an independent gating network MetaG that produces a (𝐾 + 1)-dimensional scalar weight vector g = [𝑔 (1) 𝑠ℎ , . . . , 𝑔 (𝐾) 𝑠ℎ , 𝑔𝑠𝑝𝑒𝑐 ]. For the Generalization Branch, the ga… view at source ↗
Figure 6
Figure 6. Figure 6: Dataset statistics. The bar charts (log-scale) illustrate order volume in source and target domains, while the line plots show the corresponding average ATA [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Comparison of Reliability Diagrams [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Ablation studies of UME.The y-axis shows the relative MAE degradation compared to the full UME model (lower is better). machine intelligence 45, 4 (2022), 4396–4415. [51] Lin Zhu, Wei Yu, Kairong Zhou, Xing Wang, Wenxing Feng, Pengyu Wang, Ning Chen, and Pei Lee. 2020. Order fulfillment cycle time estimation for on-demand food delivery. In Proceedings of the 26th ACM SIGKDD International Conference on Know… view at source ↗
Figure 11
Figure 11. Figure 11: Training dynamics comparison between UME with the step-wise warm-up strategy and a baseline with constant distillation (𝑇𝑤𝑎𝑟𝑚 = 0). The warm-up strategy ensures stable convergence and leads to a lower final MAE. C Sensitivity Analysis To evaluate the robustness of the knowledge distillation strategy in UME and examine the necessity of the Warm-up during early training, we conduct a sensitivity analysis on… view at source ↗
Figure 9
Figure 9. Figure 9: Reliability diagrams for the ablation study. (a) and (b) show results on the cold start target Domain. (c) and (d) show the corresponding results on the mature source domains. The diagonal line represents perfect calibration. UME consistently exhibits lower calibration error (closer to the diagonal) compared to all baselines [PITH_FULL_IMAGE:figures/full_fig_p012_9.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

55 extracted references · 4 canonical work pages · 3 internal anchors

  1. [1]

    Arjanit Arifi, Panagiotis Bouros, and Theodoros Chondrogiannis. 2024. A Study on ETA Prediction using Machine Learning and Recovered Routes.. InEDBT/ICDT Workshops

    2024

  2. [2]

    Martin Arjovsky, Léon Bottou, Ishaan Gulrajani, and David Lopez-Paz. 2019. Invariant risk minimization.arXiv preprint arXiv:1907.02893(2019)

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  3. [3]

    Xulei Cao, Wenyu Zhang, Feiyang Jiang, and Xinming Zhang. 2025. An Industrial Framework for Cold-Start Recommendation in Few-Shot and Zero-Shot Scenarios. Information16, 12 (2025), 1105

    2025

  4. [4]

    Yuting Chen, Yanshi Wang, Yabo Ni, An-Xiang Zeng, and Lanfen Lin. 2020. Scenario-aware and Mutual-based approach for Multi-scenario Recommendation in E-Commerce. In2020 International Conference on Data Mining Workshops (ICDMW). IEEE, 127–135

    2020

  5. [5]

    Zebin Chen, Xiaolin Xiao, Yue-Jiao Gong, Jun Fang, Nan Ma, Hua Chai, and Zhiguang Cao. 2022. Interpreting trajectories from multiple views: A hierarchical self-attention network for estimating the time of arrival. InProceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 2771–2779

    2022

  6. [6]

    Robert Coop. 2021. What is the Cost to Deploy and Maintain a Machine Learning Model.phData(2021). KDD ’26, August 09–13, 2026, Jeju Island, Republic of Korea Duo Wang et al

    2021

  7. [7]

    Hao Dong, Ismail Nejjar, Han Sun, Eleni Chatzi, and Olga Fink. 2023. SimM- MDG: A simple and effective framework for multi-modal domain generalization. Advances in Neural Information Processing Systems36 (2023), 78674–78695

    2023

  8. [8]

    DoorDash Engineering. 2023. Precision in Motion: Deep Learning for Smarter ETA Predictions. https://careersatdoordash.com/blog/deep-learning-for-smarter- eta-predictions/. Accessed: 2024-05-20

    2023

  9. [9]

    Tilmann Gneiting, Fadoua Balabdaoui, and Adrian E Raftery. 2007. Probabilistic forecasts, calibration and sharpness.Journal of the Royal Statistical Society Series B: Statistical Methodology69, 2 (2007), 243–268

    2007

  10. [10]

    Jindong Han, Hao Liu, Shui Liu, Xi Chen, Naiqiang Tan, Hua Chai, and Hui Xiong. 2023. iETA: A robust and scalable incremental learning framework for time-of-arrival estimation. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 4100–4111

    2023

  11. [11]

    Florentin D Hildebrandt and Marlin W Ulmer. 2022. Supervised learning for arrival time estimations in restaurant meal delivery.Transportation Science56, 4 (2022), 1058–1084

    2022

  12. [12]

    Jilin Hu, Bin Yang, Chenjuan Guo, Christian S Jensen, and Hui Xiong. 2020. Stochastic origin-destination matrix forecasting using dual-stage graph convolu- tional, recurrent neural networks. In2020 IEEE 36th International conference on data engineering (ICDE). IEEE, 1417–1428

    2020

  13. [13]

    Yilun Jin, Kai Chen, and Qiang Yang. 2022. Selective cross-city transfer learning for traffic prediction via source city region re-weighting. InProceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining. 731–741

    2022

  14. [14]

    Mahesh Joshi, Mark Dredze, William Cohen, and Carolyn Rose. 2012. Multi- domain learning: when do domains matter?. InProceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computa- tional Natural Language Learning. 1302–1312

    2012

  15. [15]

    Pengcheng Li, Runze Li, Qing Da, An-Xiang Zeng, and Lijun Zhang. 2020. Improv- ing multi-scenario learning to rank in e-commerce by exploiting task relation- ships in the label space. InProceedings of the 29th ACM International Conference on Information & Knowledge Management. 2605–2612

    2020

  16. [16]

    Maria Michalopoulou, Panayiotis Kolios, Tania Panayiotou, and Georgios Ellinas

  17. [17]

    IEEE Potentials43, 1 (2023), 20–27

    Meal delivery services: Current practices, challenges, and future directions. IEEE Potentials43, 1 (2023), 20–27

    2023

  18. [18]

    Krikamol Muandet, David Balduzzi, and Bernhard Schölkopf. 2013. Domain generalization via invariant feature representation. InInternational conference on machine learning. PMLR, 10–18

    2013

  19. [19]

    Alexandru Niculescu-Mizil and Rich Caruana. 2005. Predicting good probabilities with supervised learning. InProceedings of the 22nd international conference on Machine learning. 625–632

    2005

  20. [20]

    Aidan O’Sullivan, Francisco C Pereira, Jinhua Zhao, and Harilaos N Koutsopoulos

  21. [21]

    Uncertainty in bus arrival time predictions: Treating heteroscedasticity with a metamodel approach.IEEE Transactions on Intelligent Transportation Systems17, 11 (2016), 3286–3296

    2016

  22. [22]

    Sinno Jialin Pan and Qiang Yang. 2009. A survey on transfer learning.IEEE Transactions on knowledge and data engineering22, 10 (2009), 1345–1359

    2009

  23. [23]

    Zheyi Pan, Yuxuan Liang, Weifeng Wang, Yong Yu, Yu Zheng, and Junbo Zhang

  24. [24]

    InProceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining

    Urban traffic prediction from spatio-temporal data using deep meta learning. InProceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining. 1720–1730

  25. [25]

    Jingang Qu, Thibault Faney, Ze Wang, Patrick Gallinari, Soleiman Yousef, and Jean-Charles de Hemptinne. 2022. Hmoe: Hypernetwork-based mixture of experts for domain generalization.arXiv preprint arXiv:2211.08253(2022)

    work page arXiv 2022

  26. [26]

    Sijie Ruan, Zi Xiong, Cheng Long, Yiheng Chen, Jie Bao, Tianfu He, Ruiyuan Li, Shengnan Wu, Zhongyuan Jiang, and Yu Zheng. 2020. Doing in one go: delivery time inference based on couriers’ trajectories. InProceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2813–2821

    2020

  27. [27]

    Xiang-Rong Sheng, Liqin Zhao, Guorui Zhou, Xinyao Ding, Binding Dai, Qiang Luo, Siran Yang, Jingshan Lv, Chi Zhang, Hongbo Deng, et al. 2021. One model to serve all: Star topology adaptive recommender for multi-domain ctr prediction. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management. 4104–4113

    2021

  28. [28]

    Zijian Song, Wenhan Zhang, Lifang Deng, Jiandong Zhang, Kaigui Bian, and Bin Cui. 2024. Multilora: Multi-directional low rank adaptation for multi-domain recommendation. InProceedings of the 33rd ACM International Conference on Information and Knowledge Management. 2148–2157

    2024

  29. [29]

    Hongjie Sui, Huan Yan, Tianyi Zheng, Wenzhen Huang, Yunlin Zhuang, and Yong Li. 2024. Congestion-aware Spatio-Temporal Graph Convolutional Network- based A* Search Algorithm for Fastest Route Search.ACM Transactions on Knowledge Discovery from Data18, 7 (2024), 1–19

    2024

  30. [30]

    Hongyan Tang, Junning Liu, Ming Zhao, and Xudong Gong. 2020. Progressive layered extraction (ple): A novel multi-task learning (mtl) model for personalized recommendations. InProceedings of the 14th ACM conference on recommender systems. 269–278

    2020

  31. [31]

    Zhen Tian, Changwang Zhang, Wayne Xin Zhao, Xin Zhao, Ji-Rong Wen, and Zhao Cao. 2025. UFIN: Universal feature interaction network for multi-domain click-through rate prediction. InInternational Conference on Database Systems for Advanced Applications. Springer, 3–18

    2025

  32. [32]

    Tomer Volk, Eyal Ben-David, Ohad Amosy, Gal Chechik, and Roi Reichart. 2023. Example-based hypernetworks for multi-source adaptation to unseen domains. In Findings of the Association for Computational Linguistics: EMNLP 2023. 9096–9113

    2023

  33. [33]

    Dong Wang, Junbo Zhang, Wei Cao, Jian Li, and Yu Zheng. 2018. When will you arrive? Estimating travel time based on deep neural networks. InProceedings of the AAAI conference on artificial intelligence, Vol. 32

    2018

  34. [34]

    Jindong Wang, Cuiling Lan, Chang Liu, Yidong Ouyang, Tao Qin, Wang Lu, Yiqiang Chen, Wenjun Zeng, and Philip S Yu. 2022. Generalizing to unseen domains: A survey on domain generalization.IEEE transactions on knowledge and data engineering35, 8 (2022), 8052–8072

    2022

  35. [35]

    Leye Wang, Xu Geng, Xiaojuan Ma, Feng Liu, and Qiang Yang. 2018. Cross- city transfer learning for deep spatio-temporal prediction.arXiv preprint arXiv:1802.00386(2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  36. [36]

    Ruoxi Wang, Rakesh Shivanna, Derek Cheng, Sagar Jain, Dong Lin, Lichan Hong, and Ed Chi. 2021. Dcn v2: Improved deep & cross network and practical lessons for web-scale learning to rank systems. InProceedings of the web conference 2021. 1785–1797

    2021

  37. [37]

    Yufei Wang, Haoliang Li, and Alex C Kot. 2020. Heterogeneous domain general- ization via domain mixup. InICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 3622–3626

    2020

  38. [38]

    Zheng Wang, Kun Fu, and Jieping Ye. 2018. Learning to estimate the travel time. InProceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining. 858–866

    2018

  39. [39]

    Aasim Ayaz Wani. 2025. Ten quick tips for improving estimated time of arrival predictions using machine learning in logistics and transportation systems.PeerJ Computer Science11 (2025), e3259

    2025

  40. [40]

    Jiawen Wei, Ziwen Ye, Chuan Yang, Chen Chen, and Guangrui Ma. 2024. Process- Informed Deep Learning for Enhanced Order Fulfillment Cycle Time Prediction in On-Demand Grocery Retailing. InProceedings of the 33rd ACM International Conference on Information and Knowledge Management. 4975–4982

    2024

  41. [41]

    Shitao Xiao, Zheng Liu, Peitian Zhang, Niklas Muennighoff, Defu Lian, and Jian-Yun Nie. 2024. C-pack: Packed resources for general chinese embeddings. InProceedings of the 47th international ACM SIGIR conference on research and development in information retrieval. 641–649

    2024

  42. [42]

    Bencheng Yan, Pengjie Wang, Kai Zhang, Feng Li, Hongbo Deng, Jian Xu, and Bo Zheng. 2022. Apg: Adaptive parameter generation network for click-through rate prediction.Advances in Neural Information Processing Systems35 (2022), 24740–24752

    2022

  43. [43]

    Yongxin Yang and Timothy M Hospedales. 2014. A unified perspective on multi- domain and multi-task learning.arXiv preprint arXiv:1412.7489(2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  44. [44]

    Jinhui Yi, Huan Yan, Haotian Wang, Jian Yuan, and Yong Li. 2023. Deepsta: A spatial-temporal attention network for logistics delivery timely rate prediction in anomaly conditions. InProceedings of the 32nd ACM International Conference on Information and Knowledge Management. 4916–4922

    2023

  45. [45]

    Jinhui Yi, Huan Yan, Haotian Wang, Jian Yuan, and Yong Li. 2024. Learning to Estimate Package Delivery Time in Mixed Imbalanced Delivery and Pickup Logistics Services. InProceedings of the 32nd ACM International Conference on Advances in Geographic Information Systems. 432–443

    2024

  46. [46]

    Jinhui Yi, Huan Yan, Haotian Wang, Jian Yuan, and Yong Li. 2024. RCCNet: A Spatial-Temporal Neural Network Model for Logistics Delivery Timely Rate Prediction.ACM Transactions on Intelligent Systems and Technology15, 6 (2024), 1–21

    2024

  47. [47]

    Lei Zhang, Xin Zhou, Zhiwei Zeng, Yiming Cao, Yonghui Xu, Mingliang Wang, Xingyu Wu, Yong Liu, Lizhen Cui, and Zhiqi Shen. 2023. Delivery time prediction using large-scale graph structure learning based on quantile regression. In2023 IEEE 39th International Conference on Data Engineering (ICDE). IEEE, 3403–3416

    2023

  48. [48]

    Marvin Zhang, Henrik Marklund, Nikita Dhawan, Abhishek Gupta, Sergey Levine, and Chelsea Finn. 2021. Adaptive risk minimization: Learning to adapt to domain shift.Advances in Neural Information Processing Systems34 (2021), 23664–23678

    2021

  49. [49]

    Qianqian Zhang, Xinru Liao, Quan Liu, Jian Xu, and Bo Zheng. 2022. Leaving no one behind: A multi-scenario multi-task meta learning approach for advertiser modeling. InProceedings of the Fifteenth ACM International Conference on Web Search and Data Mining. 1368–1376

    2022

  50. [50]

    Yifei Zhang, Hua Hua, Hui Guo, Shuangyang Wang, Chongyu Zhong, and Shijie Zhang. 2023. 3mn: Three meta networks for multi-scenario and multi-task learning in online advertising recommender systems. InProceedings of the 32nd ACM International Conference on Information and Knowledge Management. 4945– 4951

    2023

  51. [51]

    Yu Zheng, Licia Capra, Ouri Wolfson, and Hai Yang. 2014. Urban computing: concepts, methodologies, and applications.ACM Transactions on Intelligent Systems and Technology (TIST)5, 3 (2014), 1–55

    2014

  52. [52]

    Jie Zhou, Xianshuai Cao, Wenhao Li, Lin Bo, Kun Zhang, Chuan Luo, and Qian Yu

  53. [53]

    In2023 IEEE 39th International Conference on Data Engineering (ICDE)

    Hinet: Novel multi-scenario & multi-task learning with hierarchical infor- mation extraction. In2023 IEEE 39th International Conference on Data Engineering (ICDE). IEEE, 2969–2975

  54. [54]

    Kaiyang Zhou, Ziwei Liu, Yu Qiao, Tao Xiang, and Chen Change Loy. 2022. Domain generalization: A survey.IEEE transactions on pattern analysis and UME: A Unified Meta-Generalization Framework for Cross-Domain ETA KDD ’26, August 09–13, 2026, Jeju Island, Republic of Korea Figure 8:Ablation studies of UME.The y-axis shows the relative MAE degradation compar...

    2022

  55. [55]

    Lin Zhu, Wei Yu, Kairong Zhou, Xing Wang, Wenxing Feng, Pengyu Wang, Ning Chen, and Pei Lee. 2020. Order fulfillment cycle time estimation for on-demand food delivery. InProceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2571–2580. A Domain Semantic Knowledge We construct a unified prompt template to standard...

    2020