Paper: arXiv 2311.18749

Authors: Jie Shi, Arno P. J. M. Siebes, Siamak Mehrkanoon

Abstract

This paper proposes an interpretable two-stream transformer CORAL networks (TransCORALNet) for supply chain credit assessment under the segment industry and cold start problem. The model aims to provide accurate credit assessment prediction for new supply chain borrowers with limited historical data. Here, the two-stream domain adaptation architecture with correlation alignment (CORAL) loss is used as a core model and is equipped with transformer, which provides insights about the learned features and allow efficient parallelization during training. Thanks to the domain adaptation capability of the proposed model, the domain shift between the source and target domain is minimized. Therefore, the model exhibits good generalization where the source and target do not follow the same distribution, and a limited amount of target labeled instances exist. Furthermore, we employ Local Interpretable Model-agnostic Explanations (LIME) to provide more insight into the model prediction and identify the key features contributing to supply chain credit assessment decisions. The proposed model addresses four significant supply chain credit assessment challenges: domain shift, cold start, imbalanced-class and interpretability. Experimental results on a real-world data set demonstrate the superiority of TransCORALNet over a number of state-of-the-art baselines in terms of accuracy. The code is available on GitHub https://github.com/JieJieNiu/TransCORALN .

Complexity vs Empirical Score

  • Math Complexity: 6.5/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper effectively combines deep learning architectures (Transformers, CORAL domain adaptation) with practical financial applications (supply chain credit cold-start). It demonstrates strong empirical rigor using real-world data and offers high reproducibility by providing an open-source GitHub repository and interpretable LIME frameworks.

Research Flowchart

  flowchart TD
    A[Goal: Credit Assessment Cold Start & Domain Shift in Supply Chains] --> B[Input: Source Domain Data & Limited Target Domain Data]
    B --> C[Process: Two-Stream Transformer Architecture]
    C --> D[Process: Domain Alignment via CORAL Loss & Class Rebalancing]
    D --> E[Process: Model Explainability via LIME]
    E --> F[Outcomes: Superior Prediction Accuracy & Feature Interpretability]

Cited References (73)

  1. As banks continue their digital transformations, they can follow four best practices for automated credit-decisioning models to incorporate more of the right data to meet future challenges (2021)
  2. Deep domain confusion: Maximizing for domain invariance (2014) — link
  3. Supply chain risk management with machine learning technology: A literature review and future research directions (2022)
  4. Fairness in credit scoring: Assessment, implementation and profit implications (2022)
  5. Domain Adaptation: Challenges, Methods, Datasets, and Applications (2023)
  6. Integrating structured biological data by kernel maximum mean discrepancy (2006)
  7. Deep CORAL: Correlation Alignment for Deep Domain Adaptation (2016)
  8. Supervised representation learning: Transfer learning with deep autoencoders (2015)
and 65 more
  1. Domain adaptation learning based on structural similarity weighted mean discrepancy for credit risk classification (2020)
  2. Attention is all you need (2017)
  3. Modeling tabular data using conditional gan (2019)
  4. Domain adaptation approach for credit risk analysis (2018)
  5. Generative adversarial networks (2020)
  6. " Why should i trust you?" Explaining the predictions of any classifier (2016)
  7. Domain-adversarial training of neural networks (2016)
  8. Conditional adversarial domain adaptation (2018)
  9. Contrastive Adaptation Network for Unsupervised Domain Adaptation (2019) — link
  10. Long short-term memory-networks for machine reading (2016) — link
  11. Sma\uppercase A t-\uppercase UN et: Precipitation nowcasting using a small attention-UNet architecture (2021)
  12. Attention Augmented Convolutional Networks (2019)
  13. Self-attention with relative position representations (2018) — link
  14. Interpretable deep learning model for building energy consumption prediction based on attention mechanism (2021)
  15. Evaluating self-attention interpretability through human-grounded experimental protocol (2023) — link
  16. Sleeptransformer: Automatic sleep staging with interpretability and uncertainty quantification (2022)
  17. \uppercase AA -\uppercase t rans\uppercase un et: Attention augmented transunet for nowcasting tasks (2022)
  18. Attention is not not explanation (2019) — link
  19. Attention is not Explanation (2019)
  20. Transfer Learning Strategies for Credit Card Fraud Detection (2021) — link
  21. Credit Risk Modelling using Transfer Learning and Domain Adaptation (2022)
  22. Oversampling method using outlier detectable generative adversarial network (2019)
  23. Ctab-gan+: Enhancing tabular data synthesis (2022) — link
  24. FCT-GAN: Enhancing Table Synthesis via Fourier Transform (2022)
  25. Conditional Wasserstein generative adversarial network-gradient penalty-based approach to alleviating imbalanced data classification (2020)
  26. CTTGAN: Traffic Data Synthesizing Scheme Based on Conditional GAN (2022)
  27. Tabfairgan: Fair tabular data generation with generative adversarial networks (2022)
  28. Applied logistic regression (2013)
  29. A random forest guided tour (2016)
  30. Support vector machines (1998)
  31. An introduction to classification and regression tree (CART) analysis (2000)
  32. Explainable Artificial Intelligence and its potential within Industry (2019)
  33. Artificial intelligence for the financial services industry: What challenges organizations to succeed (2019)
  34. Explainable AI in fintech risk management (2020)
  35. Explainable AI in credit risk management (2021) — link
  36. SHAP and LIME: an evaluation of discriminative power in credit risk (2021)
  37. Exploring explainable ai in the financial sector: Perspectives of banks and supervisory authorities (2022)
  38. Financial risk management and explainable, trustworthy, responsible AI (2022)
  39. Deep asymmetric transfer network for unbalanced domain adaptation (2018)
  40. Layer normalization (2016) — link
  41. Deep residual learning for image recognition (2016)
  42. On information and sufficiency (1951)
  43. Regularized semipaired kernel \uppercase CCA for domain adaptation (2017)
  44. Cross-domain neural-kernel networks (2019)
  45. Domain adaptation for visual applications: A comprehensive survey (2017) — link
  46. A deep domain adaptation framework with correlation alignment for EEG-based motor imagery classification (2023)
  47. Classifier inconsistency-based domain adaptation network for partial transfer intelligent diagnosis (2019)
  48. Deep-learning domain adaptation techniques for credit cards fraud detection (2020)
  49. An introductory survey on attention mechanisms in NLP problems (2020)
  50. Toward interpretable music tagging with self-attention (2019) — link
  51. Interpretable multi-head self-attention architecture for sarcasm detection in social media (2021)
  52. XAI—Explainable artificial intelligence (2019)
  53. Explainable artificial intelligence: A survey (2018) — link
  54. Model adaptation: Unsupervised domain adaptation without source data (2020)
  55. StyleGAN-NADA: CLIP-guided domain adaptation of image generators (2022)
  56. Enhancing brain decoding using attention augmented deep neural networks (2021) — link
  57. Stand-alone self-attention in vision models (2019)
  58. Fully connected deep structured networks (2015) — link
  59. A survey of loss functions for semantic segmentation (2020)
  60. A probabilistic interpretation of precision, recall and F-score, with implication for evaluation (2005)
  61. Unsupervised domain adaptation by backpropagation (2015)
  62. Inception-v4, inception-resnet and the impact of residual connections on learning (2017)
  63. The real-world-weight cross-entropy loss function: Modeling the costs of mislabeling (2019)
  64. A two-stage weighting framework for multi-source domain adaptation (2011)
  65. A multivariate Kolmogorov-Smirnov test of goodness of fit (1997)

0 of 73 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.