pith. sign in

arxiv: 2404.13004 · v6 · submitted 2024-04-19 · 💻 cs.CE · cs.AI

A Unified Framework for Modeling Heterogeneous Financial Data via Dual-Granularity Prompting

Yu Lei , Zixuan Wang , Yiqing Feng , Junru Zhang , Yahui Li , Chu Liu , Tongyao Wang , Dongyang Li This is my paper

Pith reviewed 2026-05-24 02:19 UTC · model grok-4.3

classification 💻 cs.CE cs.AI
keywords credit scoringfinancial dataprompt learningsequential modelingheterogeneous datatemporal patternsrisk predictiondeep learning
0
0 comments X

The pith

FinLangNet reformulates credit scoring as multi-scale sequential learning with dual-granularity prompts to handle heterogeneous financial data.

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

The paper establishes that a dual-module architecture combining tabular feature extraction with temporal sequence modeling, augmented by learnable prompts at both feature and user levels, can generate multi-horizon probability distributions of future financial behaviors. This addresses the persistent gap where deep learning has not reliably outperformed manually tuned statistical methods on complex, evolving credit data. A sympathetic reader would care because more accurate modeling of heterogeneous inputs directly affects lending decisions, bad debt management, and the feasibility of deploying deep models in production credit systems.

Core claim

FinLangNet processes heterogeneous financial data through a dual-module architecture that combines tabular feature extraction with temporal sequence modeling, generating probability distributions of users' future financial behaviors across multiple time horizons. A key innovation is the dual-prompt mechanism within the sequential module, which introduces learnable prompts operating at both feature-level granularity for capturing fine-grained temporal patterns and user-level granularity for aggregating holistic risk profiles. Real world deployment yielded a 6.3 pp improvement in KS, along with a 9.9% reduction in bad debt rate.

What carries the argument

Dual-prompt mechanism with learnable prompts at feature-level granularity for fine-grained temporal patterns and user-level granularity for holistic risk profiles.

If this is right

  • The model produces probability distributions of future financial behaviors across multiple time horizons rather than single-point predictions.
  • Tabular feature extraction and temporal sequence modeling together address the heterogeneity that has limited prior deep learning approaches.
  • Feature-level prompts capture fine-grained temporal patterns while user-level prompts aggregate overall risk.
  • Deployment results indicate measurable gains in KS metric and bad debt reduction under industrial conditions.

Where Pith is reading between the lines

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

  • The multi-scale framing could be tested on other mixed tabular-sequential domains such as transaction forecasting or user lifetime value prediction.
  • The separation of prompt granularities suggests a way to balance local pattern detection with global profile stability in any evolving user modeling task.
  • If the prompting mechanism generalizes, it may reduce the need for extensive manual feature engineering in financial risk systems.

Load-bearing premise

The observed performance gains are caused by the dual-granularity prompting mechanism rather than other unstated factors such as data preprocessing choices, deployment-specific tuning, or differences in the evaluation environment.

What would settle it

An ablation experiment on the same real-world dataset that removes only the dual-prompt mechanism while holding all other components and preprocessing fixed, then measures whether the KS improvement and bad debt reduction disappear.

Figures

Figures reproduced from arXiv: 2404.13004 by Chu Liu, Dongyang Li, Junru Zhang, Tongyao Wang, Yahui Li, Yiqing Feng, Yu Lei, Zixuan Wang.

Figure 1
Figure 1. Figure 1: The dataset consists of two parts: Non-Sequential Data [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FinLangNet Framework Overview. The architecture incorporates two pivotal sub-modules to harness both sequential and non [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Model Performance on the UEA archive. (i.e., using only the SRG module), our method achieves state-of-the￾art performance across all five datasets. The accuracy comparison results, illustrated in [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Overview of Deployment. (a) Online deployment and data flow. (b) Default [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Performance comparison at various risk thresholds for [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Distribution of predicted risk scores versus actual [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
read the original abstract

Recent industrial credit scoring models remain heavily reliant on manually tuned statistical learning methods. Despite their potential, deep learning architectures have struggled to consistently outperform traditional statistical models in industrial credit scoring, largely due to the complexity of heterogeneous financial data and the challenge of modeling evolving creditworthiness. To bridge this gap, we introduce FinLangNet, a novel framework that reformulates credit scoring as a multi-scale sequential learning problem. FinLangNet processes heterogeneous financial data through a dual-module architecture that combines tabular feature extraction with temporal sequence modeling, generating probability distributions of users' future financial behaviors across multiple time horizons. A key innovation is our dual-prompt mechanism within the sequential module, which introduces learnable prompts operating at both feature-level granularity for capturing fine-grained temporal patterns and user-level granularity for aggregating holistic risk profiles. Notably, real world deployment yielded a 6.3 pp improvement in KS, along with a 9.9\% reduction in bad debt rate.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces FinLangNet, a framework that reformulates credit scoring as multi-scale sequential learning on heterogeneous financial data. It uses a dual-module architecture (tabular feature extraction plus temporal sequence modeling) and proposes a dual-granularity prompting mechanism (feature-level and user-level learnable prompts) to generate probability distributions over future financial behaviors. The central empirical claim is that real-world deployment produced a 6.3 pp lift in KS and a 9.9% reduction in bad-debt rate.

Significance. If the deployment results can be shown to be causally attributable to the dual-granularity prompting rather than to unmeasured factors, the work would offer a concrete path for deep-learning methods to outperform manually tuned statistical models in industrial credit scoring, addressing long-standing difficulties with data heterogeneity and evolving creditworthiness.

major comments (2)
  1. [Abstract] Abstract: the reported deployment gains (6.3 pp KS, 9.9% bad-debt reduction) are presented without any description of the baseline production model, the experimental design (A/B test versus before/after), the exact definition or computation of KS in the live environment, or controls for concurrent changes in feature engineering, data pipelines, or hyper-parameter search. This omission renders the central claim that the gains are due to the dual-prompt mechanism unverifiable.
  2. [Abstract] Abstract / deployment results paragraph: because no information is supplied on how the evaluation environment was held constant or how the baseline was chosen, the observed deltas cannot be isolated from other unstated factors, directly undermining the load-bearing empirical assertion.
minor comments (1)
  1. [Abstract] Abstract: the acronym 'KS' is used without expansion or reference on first appearance.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for highlighting the need for greater transparency around the deployment results. We agree that the current abstract provides insufficient detail to allow readers to assess the experimental controls and isolate the contribution of the dual-granularity prompting mechanism. We address the two major comments below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the reported deployment gains (6.3 pp KS, 9.9% bad-debt reduction) are presented without any description of the baseline production model, the experimental design (A/B test versus before/after), the exact definition or computation of KS in the live environment, or controls for concurrent changes in feature engineering, data pipelines, or hyper-parameter search. This omission renders the central claim that the gains are due to the dual-prompt mechanism unverifiable.

    Authors: We agree that the abstract as written does not supply enough information for independent verification. In the revised version we will expand the abstract and add a dedicated paragraph in the Experiments section that states: (i) the evaluation was performed as a controlled online A/B test in which the treatment arm used FinLangNet while the control arm used the existing production model; (ii) KS was computed in the standard manner as the supremum of the absolute difference between the empirical CDFs of scores for default and non-default accounts; and (iii) feature engineering, data pipelines, and hyper-parameters were frozen for the duration of the test. Because of commercial confidentiality and data-protection regulations we cannot disclose the precise architecture or feature set of the baseline production model; we will explicitly note this limitation. revision: yes

  2. Referee: [Abstract] Abstract / deployment results paragraph: because no information is supplied on how the evaluation environment was held constant or how the baseline was chosen, the observed deltas cannot be isolated from other unstated factors, directly undermining the load-bearing empirical assertion.

    Authors: We acknowledge the validity of this concern. The revision described above will make explicit that the A/B test held the data pipeline, feature set, and scoring threshold constant, with the sole change being the replacement of the production scorer by FinLangNet. This design isolates the model change to the greatest extent feasible under operational constraints. We will also add a short limitations paragraph stating that, while concurrent changes were minimized, complete isolation from all external factors cannot be guaranteed in a live production environment. revision: yes

standing simulated objections not resolved
  • Exact architecture and feature composition of the proprietary baseline production model

Circularity Check

0 steps flagged

No circularity; empirical deployment results with no derivation chain

full rationale

The paper introduces FinLangNet as a framework reformulating credit scoring via dual-module architecture and dual-prompt mechanism, then reports real-world deployment metrics (6.3 pp KS improvement, 9.9% bad debt reduction). No equations, predictions, or first-principles derivations are presented that could reduce to inputs by construction. The central claims are empirical outcomes rather than derived results, so none of the enumerated circularity patterns apply.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract describes an applied machine learning system without mathematical derivations, fitted constants, or new postulated entities.

pith-pipeline@v0.9.0 · 5712 in / 1110 out tokens · 26892 ms · 2026-05-24T02:19:24.337337+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

65 extracted references · 65 canonical work pages · 7 internal anchors

  1. [1]

    Behrouz Ahadzadeh, Moloud Abdar, Fatemeh Safara, Abbas Khosravi, Moham- mad Bagher Menhaj, and Ponnuthurai Nagaratnam Suganthan. 2023. SFE: A simple, fast and efficient feature selection algorithm for high-dimensional data. IEEE Transactions on Evolutionary Computation(2023)

    work page 2023

  2. [2]

    Maher Ala’raj, Maysam F Abbod, and Munir Majdalawieh. 2021. Modelling customers credit card behaviour using bidirectional LSTM neural networks. Journal of Big Data8, 1 (2021), 69

    work page 2021

  3. [3]

    Anthony Bagnall, Hoang Anh Dau, Jason Lines, Michael Flynn, James Large, Aaron Bostrom, Paul Southam, and Eamonn Keogh. 2018. The UEA multivariate time series classification archive, 2018.arXiv preprint arXiv:1811.00075(2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  4. [4]

    Vicente Balmaseda, María Coronado, and Gonzalo de Cadenas-Santiago. 2023. Predicting systemic risk in financial systems using deep graph learning.Intelligent Systems with Applications19 (2023), 200240

    work page 2023

  5. [5]

    Vadim Borisov, Tobias Leemann, Kathrin Seßler, Johannes Haug, Martin Pawel- czyk, and Gjergji Kasneci. 2022. Deep neural networks and tabular data: A survey. IEEE Transactions on Neural Networks and Learning Systems(2022)

    work page 2022

  6. [6]

    Michael Bücker, Gero Szepannek, Alicja Gosiewska, and Przemyslaw Biecek

    work page

  7. [7]

    Transparency, auditability, and explainability of machine learning models in credit scoring.Journal of the Operational Research Society73, 1 (2022), 70–90

    work page 2022

  8. [8]

    Weichen Chen, Canru Wang, Yaming Wang, Wei Liu, Ying Lyu, Taifeng Chen, and Yong Yu. 2019. Behavior Sequence Transformer for E-commerce Recommendation in Alibaba. InProceedings of the 1st Workshop on Deep Learning Practice for High- Dimensional Sparse Data. ACM, 1–4

    work page 2019

  9. [9]

    Dawei Cheng, Zhibin Niu, and Yiyi Zhang. 2020. Contagious chain risk rating for networked-guarantee loans. InProceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2715–2723

    work page 2020

  10. [10]

    Kyunghyun Cho, Bart van Merrienboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. 2014. Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation. arXiv preprint arXiv:1406.1078(2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  11. [11]

    Angus Dempster, François Petitjean, and Geoffrey I Webb. 2020. ROCKET: excep- tionally fast and accurate time series classification using random convolutional kernels.Data Mining and Knowledge Discovery34, 5 (2020), 1454–1495

    work page 2020

  12. [12]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers). 4171–4186

    work page 2019

  13. [13]

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xi- aohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929(2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  14. [14]

    Ayoub El-Qadi, Maria Trocan, Thomas Frossard, and Natalia Díaz-Rodríguez

    work page

  15. [15]

    InPhysical Sciences Forum, Vol

    Credit Risk Scoring Forecasting Using a Time Series Approach. InPhysical Sciences Forum, Vol. 5. MDPI, 16

    work page

  16. [16]

    Gianluca Elia, Valeria Stefanelli, and Greta Benedetta Ferilli. 2023. Investigating the role of Fintech in the banking industry: what do we know?European Journal of Innovation Management26, 5 (2023), 1365–1393

    work page 2023

  17. [17]

    Jacob Fein-Ashley, Tian Ye, Sachini Wickramasinghe, Bingyi Zhang, Rajgopal Kannan, and Viktor Prasanna. 2024. A single graph convolution is all you need: Efficient grayscale image classification.arXiv preprint arXiv:2402.00564(2024)

    work page arXiv 2024

  18. [18]

    Javad Forough and Saeedeh Momtazi. 2022. Sequential credit card fraud detection: A joint deep neural network and probabilistic graphical model approach.Expert Systems39, 1 (2022), e12795

    work page 2022

  19. [19]

    Sergio Genovesi, Julia Maria Mönig, Anna Schmitz, Maximilian Poretschkin, Maram Akila, Manoj Kahdan, Romina Kleiner, Lena Krieger, and Alexander Zim- mermann. 2023. Standardizing fairness-evaluation procedures: interdisciplinary insights on machine learning algorithms in creditworthiness assessments for small personal loans.AI and Ethics(2023), 1–17

    work page 2023

  20. [20]

    2016.Deep Learning

    Ian Goodfellow, Yoshua Bengio, and Aaron Courville. 2016.Deep Learning. MIT Press

    work page 2016

  21. [21]

    Yury Gorishniy, Ivan Rubachev, Valentin Khrulkov, and Artem Babenko. 2021. Revisiting deep learning models for tabular data.Advances in Neural Information Processing Systems34 (2021), 18932–18943

    work page 2021

  22. [22]

    Albert Gu and Tri Dao. 2023. Mamba: Linear-time sequence modeling with selective state spaces.arXiv preprint arXiv:2312.00752(2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  23. [23]

    Huifeng Guo, Ruiming Tang, Yunming Ye, Zhenguo Li, and Xiuqiang He. 2017. DeepFM: a factorization-machine based neural network for CTR prediction.arXiv preprint arXiv:1703.04247(2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  24. [24]

    Van-Sang Ha, Dang-Nhac Lu, Gyoo Seok Choi, Ha-Nam Nguyen, and Byeongnam Yoon. 2019. Improving credit risk prediction in online peer-to-peer (P2P) lending using feature selection with deep learning. In2019 21st International Conference on Advanced Communication Technology (ICACT). IEEE, 511–515

    work page 2019

  25. [25]

    Hamilton, Rex Ying, and Jure Leskovec

    William L. Hamilton, Rex Ying, and Jure Leskovec. 2017. Inductive Representation Learning on Large Graphs. InAdvances in Neural Information Processing Systems, Vol. 30

    work page 2017

  26. [26]

    Hongliang He, Wenyu Zhang, and Shuai Zhang. 2018. A novel ensemble method for credit scoring: Adaption of different imbalance ratios.Expert Systems with Applications98 (2018), 105–117

    work page 2018

  27. [27]

    Fatima Zahra Janane, Tayeb Ouaderhman, and Hasna Chamlal. 2023. A filter fea- ture selection for high-dimensional data.Journal of Algorithms & Computational Technology17 (2023), 17483026231184171

    work page 2023

  28. [28]

    Herbert L Jensen. 1992. Using neural networks for credit scoring.Managerial finance18, 6 (1992), 15–26

    work page 1992

  29. [29]

    Fazle Karim, Somshubra Majumdar, Houshang Darabi, and Samuel Harford. 2019. Multivariate LSTM-FCNs for time series classification.Neural networks116 (2019), 237–245

    work page 2019

  30. [30]

    Xuan-May Le, Ling Luo, Uwe Aickelin, and Minh-Tuan Tran. 2024. ShapeFormer: Shapelet Transformer for Multivariate Time Series Classification.arXiv preprint arXiv:2405.14608(2024)

    work page arXiv 2024

  31. [31]

    Yu Lei, Zixuan Wang, Chu Liu, and Tongyao Wang. 2025. Zigong 1.0: A large language model for financial credit.arXiv preprint arXiv:2502.16159(2025)

    work page arXiv 2025

  32. [32]

    Yu Lei, Jiayang Zhao, Yilei Zhao, Zhaoqi Zhang, Linyou Cai, Qianlong Xie, and Xingxing Wang. 2025. Generative Large-Scale Pre-trained Models for Automated Ad Bidding Optimization.arXiv preprint arXiv:2508.02002(2025)

    work page arXiv 2025

  33. [33]

    Guozhong Li, Byron Choi, Jianliang Xu, Sourav S Bhowmick, Kwok-Pan Chun, and Grace Lai-Hung Wong. 2021. Shapenet: A shapelet-neural network approach for multivariate time series classification. InProceedings of the AAAI conference on artificial intelligence, Vol. 35. 8375–8383

    work page 2021

  34. [34]

    Wei Li, Shuai Ding, Hao Wang, Yi Chen, and Shanlin Yang. 2020. Heterogeneous ensemble learning with feature engineering for default prediction in peer-to-peer lending in China.World Wide Web23 (2020), 23–45

    work page 2020

  35. [35]

    Yixuan Li, Charalampos Stasinakis, and Wee Meng Yeo. 2022. A hybrid XGBoost- MLP model for credit risk assessment on digital supply chain finance.Forecasting 4, 1 (2022), 184–207

    work page 2022

  36. [36]

    Yancheng Liang, Jiajie Zhang, Hui Li, Xiaochen Liu, Yi Hu, Yong Wu, Jinyao Zhang, Yongyan Liu, and Yi Wu. 2023. DeRisk: An Effective Deep Learning Framework for Credit Risk Prediction over Real-World Financial Data.arXiv preprint arXiv:2308.03704(2023)

    work page arXiv 2023

  37. [37]

    Xinyan Liu. 2022. Fast Recommender System Combining Global and Local Infor- mation: Construction of large-scale commodity information recommendation system. In2022 2nd International Conference on Big Data, Artificial Intelligence and Risk Management (ICBAR). IEEE, 166–169

    work page 2022

  38. [38]

    Tian Lu and Yingjie Zhang. 2023. Profit vs. Equality? The Case of Financial Risk Assessment and A New Perspective on Alternative Data.MIS Quarterly47, 4 (2023)

    work page 2023

  39. [39]

    Xiaojun Ma, Jinglan Sha, Dehua Wang, Yuanbo Yu, Qian Yang, and Xueqi Niu

    work page

  40. [40]

    Study on a prediction of P2P network loan default based on the machine learning LightGBM and XGboost algorithms according to different high dimen- sional data cleaning.Electronic Commerce Research and Applications31 (2018), 24–39

    work page 2018

  41. [41]

    Stephen E Maiden and Stephen E Maiden. 2024. FICO Score.Darden Business Publishing Cases(2024), 1–11

    work page 2024

  42. [42]

    Frank J Massey Jr. 1951. The Kolmogorov-Smirnov test for goodness of fit.Journal of the American statistical Association46, 253 (1951), 68–78

    work page 1951

  43. [43]

    Majid Niazkar, Andrea Menapace, Bruno Brentan, Reza Piraei, David Jimenez, Pranav Dhawan, and Maurizio Righetti. 2024. Applications of XGBoost in water resources engineering: A systematic literature review (Dec 2018–May 2023). Environmental Modelling & Software(2024), 105971

    work page 2024

  44. [44]

    Aniruddh Raghu, Payal Chandak, Ridwan Alam, John Guttag, and Collin Stultz

    work page

  45. [45]

    InInternational Conference on Machine Learning

    Sequential multi-dimensional self-supervised learning for clinical time series. InInternational Conference on Machine Learning. PMLR, 28531–28548

    work page

  46. [46]

    Why Should I Trust You?

    Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. 2016. "Why Should I Trust You?": Explaining the Predictions of Any Classifier. InProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, 1135–1144. https://doi.org/10.1145/2939672.2939778

    work page doi:10.1145/2939672.2939778 2016

  47. [47]

    Patrick Schäfer and Ulf Leser. 2017. Multivariate time series classification with WEASEL+ MUSE.arXiv preprint arXiv:1711.11343(2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  48. [48]

    Yu Song, Yuyan Wang, Xin Ye, Russell Zaretzki, and Chuanren Liu. 2023. Loan default prediction using a credit rating-specific and multi-objective ensemble learning scheme.Information Sciences629 (2023), 599–617

    work page 2023

  49. [49]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need.Advances in neural information processing systems30 (2017)

    work page 2017

  50. [50]

    Chongren Wang and Zhuoyi Xiao. 2022. A deep learning approach for credit scoring using feature embedded Transformer.Applied Sciences12, 21 (2022), 10995

    work page 2022

  51. [51]

    Benjamin Warner, Antoine Chaffin, Benjamin Clavié, Orion Weller, Oskar Hall- ström, Said Taghadouini, Alexis Gallagher, Raja Biswas, Faisal Ladhak, Tom Aarsen, et al. 2024. Smarter, better, faster, longer: A modern bidirectional encoder for fast, memory efficient, and long context finetuning and inference.arXiv preprint arXiv:2412.13663(2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  52. [52]

    Haixu Wu, Zonghan Xu, Yuxuan Liu, Yihan Wu, Jing Lin, Li Zeng, and Xing Xie. 2023. TimesNet: Temporal 2D-Variation Modeling for General Time Series Preprint, April 4, 2025 Yu Lei, et al. Analysis.International Conference on Learning Representations (ICLR)(2023)

    work page 2023

  53. [53]

    Junhui Xu, Zekai Lu, and Ying Xie. 2021. Loan default prediction of Chinese P2P market: a machine learning methodology.Scientific Reports11, 1 (2021), 18759

    work page 2021

  54. [54]

    Jinxin Xu, Han Wang, Yuqiang Zhong, Lichen Qin, and Qishuo Cheng. 2024. Pre- dict and Optimize Financial Services Risk Using AI-driven Technology.Academic Journal of Science and Technology10, 1 (2024), 299–304

    work page 2024

  55. [55]

    Gang Xue, Shifeng Liu, Long Ren, and Daqing Gong. 2024. Risk assessment of util- ity tunnels through risk interaction-based deep learning.Reliability Engineering & System Safety241 (2024), 109626

    work page 2024

  56. [56]

    Yuantao Yao, Minghan Yang, Jianye Wang, and Min Xie. 2022. Multivariate time- series prediction in industrial processes via a deep hybrid network under data uncertainty.IEEE Transactions on Industrial Informatics19, 2 (2022), 1977–1987

    work page 2022

  57. [57]

    Samuel Yousefi and Babak Mohamadpour Tosarkani. 2022. The adoption of new technologies for sustainable risk management in logistics planning: A sequential dynamic approach.Computers & Industrial Engineering173 (2022), 108627

    work page 2022

  58. [58]

    Yong Yu, Xiaosheng Si, Changhua Hu, and Jianxun Zhang. 2019. A review of recurrent neural networks: LSTM cells and network architectures.Neural computation31, 7 (2019), 1235–1270

    work page 2019

  59. [59]

    George Zerveas, Srideepika Jayaraman, Dhaval Patel, Anuradha Bhamidipaty, and Carsten Eickhoff. 2021. A transformer-based framework for multivariate time series representation learning. InProceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining. 2114–2124

    work page 2021

  60. [60]

    Junru Zhang, Lang Feng, Xu Guo, Yuhan Wu, Yabo Dong, and Duanqing Xu. 2025. TimeMaster: Training Time-Series Multimodal LLMs to Reason via Reinforcement Learning.arXiv preprint arXiv:2506.13705(2025)

    work page arXiv 2025

  61. [61]

    Xuchao Zhang, Yifeng Gao, Jessica Lin, and Chang-Tien Lu. 2020. Tapnet: Multi- variate time series classification with attentional prototypical network. InPro- ceedings of the AAAI conference on artificial intelligence, Vol. 34. 6845–6852

    work page 2020

  62. [62]

    Yang Zhao, John W Goodell, Yong Wang, and Mohammad Zoynul Abedin. 2023. Fintech, macroprudential policies and bank risk: Evidence from China.Interna- tional Review of Financial Analysis87 (2023), 102648

    work page 2023

  63. [63]

    Guorui Zhou, Xiao Ma, Xiaoqiang Zhu, Yingqiang Fan, Yuan Wang, Yanghui Yan, Junqi Jin, Han Li, and Kun Gai. 2019. Deep Interest Evolution Network for Click-Through Rate Prediction. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 33. 5941–5948

    work page 2019

  64. [64]

    Guorui Zhou, Xiaoqiang Zhu, Chenru Song, Yingqiang Fan, Xiao Ma, Yanghui Yan, Junqi Jin, Han Li, and Kun Gai. 2018. Deep Interest Network for Click- Through Rate Prediction. InProceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining

    work page 2018

  65. [65]

    Rundong Zuo, Guozhong Li, Byron Choi, Sourav S Bhowmick, Daphne Ngar- yin Mah, and Grace LH Wong. 2023. SVP-T: a shape-level variable-position transformer for multivariate time series classification. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 37. 11497–11505. A GENAI USAGE DISCLOSURE This paper is entirely original and authored ...

    work page arXiv 2023