臺灣博碩士論文加值系統

隨著機器學習技術在各行各業中的普及，學者和專家不斷尋求提高預測準確性的方法。集成學習(Ensemble learning) 作為一種已被證明的策略，能有效地提升模型的預測性能。其核心觀念是多個模型的預測整合後往往能夠超越單一模型的能力。本研究主要關注: 如何正確的整合這些模型的預測結果，尤其是考慮到每一位專家在不同的情境或數據上的表現。
本研究提出一種新的策略：通過對每個模型的不確定性和錯誤率進行映射和分析，製作出不確定性和錯誤率的映射表，透過對每個模型的映射表，我們可以更精確地識別哪些專家在特定情境下更具信心度和準確性，這不僅可以幫助我們選擇和整合最有信心的預測結果，而且還可以進一步提升整體預測效能。
本研究在組合零次學習(Compositional Zero-Shot Learning)和集成學習的真實資料集中進行模擬，旨在驗證我們的方法如何提升在這些資料籍上面的性能。

As machine learning technologies proliferate across various industries, scholars and experts continually seek methods to enhance prediction accuracy. Ensemble learning, a proven strategy, can effectively improve the predictive performance of models. The core idea is that the combined predictions of multiple models often surpass the capabilities of a single model. This research primarily focuses on: how to correctly integrate these models' predictions, especially considering the performance of each expert under different scenarios or data.
This study introduces a new strategy: by mapping and analyzing the uncertainty and error rates of each model, a mapping table of uncertainty and error rates is constructed. Through each model's mapping table, we can more precisely identify which experts have greater confidence and accuracy under specific circumstances. This not only assists us in selecting and integrating the most confident predictions but also further enhances the overall predictive efficacy.
This research conducts simulations on real datasets of Compositional Zero-Shot Learning(CZSL) and Ensemble Learning to verify how our method improves performance on these datasets.

Table of Contents
摘要 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Goal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Related Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1 Ensemble Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Compositional Zero-Shot Learning(CZSL) . . . . . . . . . . . . . . . . . . . . 5
2.3 Normalized Uncertainty estimation . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Uncertainty Calibration Error (UCE) . . . . . . . . . . . . . . . . . . . . . . . 6
3 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.1 Model Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.2 Model Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3 Our method: Hierarchical MOE(HMOE) . . . . . . . . . . . . . . . . . . . . . 13
3.4 Evaluating the Variance of the Average Error Rate . . . . . . . . . . . . . . . . 15
3.5 Object-attribute paired logit Estimation Architecture . . . . . . . . . . . . . . 16
4 Experiment and Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.2 Evaluation Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.3 Experiment Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
iii
4.4 SPE Uncertainty to Error Rates Mapping reliability diagram . . . . . . . . . . 24
4.5 Experiment (I): SPE with Dynamic Thresholding . . . . . . . . . . . . . . . . 27
4.6 Experiment (II): Comparison on Other Ensemble Learning Methods . . . . . . 29
4.7 Experiment (III): SPE Performance Analysis . . . . . . . . . . . . . . . . . . . 31
4.8 Ablation Study (I): SPE Outlier Estimation via POE Implementation . . . . . . 31
4.9 Ablation study(II):SPE Choose the lowest error rate . . . . . . . . . . . . . . . 33
4.10 Ablation study(III):Hierarchical MOE(HMOE) . . . . . . . . . . . . . . . . . 35
4.11 Ablation study(IV):The Effect of Model Calibration . . . . . . . . . . . . . . . 36
5 Conclusions and Future Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

References
[1] P. Isola, J. J. Lim, and E. H. Adelson, “Discovering states and transformations in image
collections,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2015, pp. 1383–1391.
[2] A. Yu and K. Grauman, “Fine-grained visual comparisons with local learning,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2014, pp.
192–199.
[3] L. Breiman, “Bagging predictors,” Machine learning, vol. 24, pp. 123–140, 1996.
[4] Y. Freund, R. E. Schapire et al., “Experiments with a new boosting algorithm,” in icml,
vol. 96. Citeseer, 1996, pp. 148–156.
[5] Y. Freund and R. E. Schapire, “A decision-theoretic generalization of on-line learning and
an application to boosting,” Journal of computer and system sciences, vol. 55, no. 1, pp.
119–139, 1997.
[6] R. A. Jacobs, M. I. Jordan, S. J. Nowlan, and G. E. Hinton, “Adaptive mixtures of local
experts,” Neural computation, vol. 3, no. 1, pp. 79–87, 1991.
[7] G. E. Hinton, “Training products of experts by minimizing contrastive divergence,” Neural
computation, vol. 14, no. 8, pp. 1771–1800, 2002.
[8] R. Polikar, “Ensemble learning,” Ensemble machine learning: Methods and applications,
pp. 1–34, 2012.
[9] E. S. Aimar, A. Jonnarth, M. Felsberg, and M. Kuhlmann, “Balanced product of experts
for long-tailed recognition,” arXiv preprint arXiv:2206.05260, 2022.
[10] D. D. Hoffman and W. A. Richards, “Parts of recognition,” Cognition, vol. 18, no. 1-3, pp.
65–96, 1984.
44
[11] C. H. Lampert, H. Nickisch, and S. Harmeling, “Learning to detect unseen object classes by
between-class attribute transfer,” in 2009 IEEE conference on computer vision and pattern
recognition. IEEE, 2009, pp. 951–958.
[12] I. Misra, A. Gupta, and M. Hebert, “From red wine to red tomato: Composition with
context,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 1792–1801.
[13] N. Saini, K. Pham, and A. Shrivastava, “Disentangling visual embeddings for attributes
and objects,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern
Recognition, 2022, pp. 13 658–13 667.
[14] S. Kumar, A. Iftekhar, E. Prashnani, and B. Manjunath, “Locl: Learning object-attribute
composition using localization,” arXiv preprint arXiv:2210.03780, 2022.
[15] X. Li, X. Yang, K. Wei, C. Deng, and M. Yang, “Siamese contrastive embedding network
for compositional zero-shot learning,” in Proceedings of the IEEE/CVF conference on
computer vision and pattern recognition, 2022, pp. 9326–9335.
[16] X. Lu, S. Guo, Z. Liu, and J. Guo, “Decomposed soft prompt guided fusion enhancing
for compositional zero-shot learning,” in Proceedings of the IEEE/CVF Conference on
Computer Vision and Pattern Recognition, 2023, pp. 23 560–23 569.
[17] C. Wang, C. Lawrence, and M. Niepert, “Uncertainty estimation and calibration with finitestate probabilistic rnns,” arXiv preprint arXiv:2011.12010, 2020.
[18] M. Huang and Y. Qiao, “Uncertainty-estimation with normalized logits for out-ofdistribution detection,” in International Conference on Computer, Artificial Intelligence,
and Control Engineering (CAICE 2023), vol. 12645. SPIE, 2023, pp. 524–530.
[19] C. Chen, A. Seff, A. Kornhauser, and J. Xiao, “Deepdriving: Learning affordance for direct
perception in autonomous driving,” in Proceedings of the IEEE international conference
on computer vision, 2015, pp. 2722–2730.
45
[20] A. Esteva, B. Kuprel, R. A. Novoa, J. Ko, S. M. Swetter, H. M. Blau, and S. Thrun,
“Dermatologist-level classification of skin cancer with deep neural networks,” nature, vol.
542, no. 7639, pp. 115–118, 2017.
[21] M.-H. Laves, S. Ihler, K.-P. Kortmann, and T. Ortmaier, “Uncertainty calibration error: A
new metric for multi-class classification,” 2020.
[22] A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell,
P. Mishkin, J. Clark et al., “Learning transferable visual models from natural language
supervision,” in International conference on machine learning. PMLR, 2021, pp. 8748–
8763.
[23] A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly et al., “An image is worth 16x16 words: Transformers for image recognition at scale,” arXiv preprint arXiv:2010.11929, 2020.