臺灣博碩士論文加值系統

機器學習技術它在近幾年來引起了很多的關注，它來越來越受歡迎並得到廣泛
應用。然而，在我們現實的應用當中，我們通常都可以獲得大量的數據，但是這
些數據大多都是未標記的數據。導致，一些我們經常使用的許多經典演算法無法
直接使用。由於學習一個好的預測模型需要大量的標籤信息，但獲取標籤信息通
常都是很難或昂貴，需要相當大的成本。因此，我們想要使用主動學習，這種算
法可以盡可能地減少標記資料以減少標記成本，但是是在仍然能訓練出一個足夠
好的預測模型的前提下。並且，我們嘗試結合了遷移學習來解決主動學習算法的
缺點：初始點選擇，並且嘗試加強主動學習的初始效能。在本篇論文中，我們提
出了一個簡單的主動學習框架結合了跨領域遷移，它可以使用來自不同任務但相
關的標記數據來提高目標任務的預測表現，在沒有初始資料的情況下。最後，我
們使用了幾個遷移學習常用的數據集來確認我們的方法，而這個實驗結果驗證了
我們提出的方法的有效性。

Machine learning technologies has attracted a lot of attention, it is becoming pop-
ular and be widely applied in most recent years. In the real world applications,
we can get a huge amount of data, but these data are unlabeled data. However,
many classic classication algorithms we often used cannot be used directly. Since
learning a good classier usually need large quantities of labels information, but get
labels information is usually dicult or expensive (need time and money). Even if
we just labeled some of the training data, the time and money cost of labeling data is
unimaginable. Therefore, we used active learning, an algorithm that can reduces the
training set and labeling cost as much as possible. And combine transfer learning
to solve the weaknesses of active learning algorithms: initial selection. Moreover,
we also try to improve the performance of active learning in the beginning. In this
paper, we propose a simple active learning framework with cross domain transfer,
which using labeled data from a dierent (but related) tasks to improve the perfor-
mance of an active learner. We use some commonly used transfer learning data
sets to conrm our method analysis. Moreover, the results of experiment verify the
eectiveness of the method we proposed.

Contents
1 Introduction 6
2 Related Work 8
3 Pre-knowledge 10
3.1 Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2 Problem setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4 Methodology 11
4.1 Active learning framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.1.1 Initial stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.1.2 Iterative process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.2 Transfer learning framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.2.1 Dimension Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.2.2 Marginal Distribution Adaptation . . . . . . . . . . . . . . . . . . . . . . . . 13
4.2.3 Conditional Distribution Adaptation . . . . . . . . . . . . . . . . . . . . . . 14
4.3 Combine two frameworks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5 Experiments 15
5.1 Data preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.2 Baseline method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
5.3 Evaluations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
5.3.1 Homogeneous Domain Adaptation . . . . . . . . . . . . . . . . . . . . . . . . 17
5.3.2 Heterogeneous Domain Adaptation . . . . . . . . . . . . . . . . . . . . . . . 17
6 Conclusion and Future Work 20
6.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
6.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

List of Figures
1.1 An example of active learning. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 The eect of the initial selection on active learning. . . . . . . . . . . . . . . . . . . 7
4.1 Process diagram of the active learning algorithm. . . . . . . . . . . . . . . . . . . . 12
4.2 An example of margin sampling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
5.1 Example images of Caltech-256 and Oce dataset. . . . . . . . . . . . . . . . . . . 16
5.2 Overview of baseline methods. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.3 The results of Homogeneous Domain Adaptation with TALt. . . . . . . . . . . . . . 18
5.4 The results of Homogeneous Domain Adaptation with TALs. . . . . . . . . . . . . . 18
5.5 The results of Heterogeneous Domain Adaptation with TALt. . . . . . . . . . . . . 19
5.6 The results of Heterogeneous Domain Adaptation with TALs. . . . . . . . . . . . . 19
5.7 The category results of Experiment. . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

References
[1] X. Wu, V. Kumar, J.R. Quinlan, J. Ghosh, Q. Yang, H. Motoda, G.J. McLachlan, A.F.M. Ng, B. Liu, P.S. Yu,
Z.-H. Zhou, M. Steinbach, D.J. Hand, and D. Steinberg, Top 10 Algorithms in Data Mining., Knowledge and
Information Systems, vol. 14, no. 1, pp. 1-37, 2008.
[2] T. Hastie, R. Tibshirani, and J. Friedman, The Elements of Statistical Learning: Data Mining, Inference, and
Prediction., Springer, 2009.
[3] I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning., MIT Press, 2016.
[4] S. C. Hoi, R. Jin, J. Zhu, M. R. Lyu, Batch mode active learning and its application to medical image classification., International Conference on Machine Learning (ICML), 2006.
[5] P. Zhao and S. C. Hoi, Cost-sensitive online active learning with application to malicious URL detection.,
SIGKDD International Conference on Knowledge Discovery and Data Mining, 2013.
[6] D. Lewis and J. Catlett, Heterogeneous uncertainty sampling for supervised learning., International Conference
on Machine Learning (ICML), 1994
[7] S. Tong and D. Koller, Support vector machine active learning with applications to text classification., Journal
of Machine Learning Research (JMLR), 2001.
[8] M. Liu, W. Buntine and G. Haffari, Learning How to Actively Learn: A Deep Imitation Learning Approach.,
Annual Meeting of the Association for Computational Linguistics (ACL), 2018.
[9] M. Fang, Y. Li, T. Cohn, Learning how to Active Learn: A Deep Reinforcement Learning Approach., Conference
on Empirical Methods in Natural Language Processing, 2017.
21
[10] H.-M. Chu and H.-T. Lin, Can active learning experience be transferred?, IEEE International Conference on
Data Mining (ICDM), 2016.
[11] K.-H. Huang and H.-T. Lin, A novel uncertainty sampling algorithm for cost-sensitive multiclass active learning.,
IEEE International Conference on Data Mining (ICDM), 2016..
[12] B. Du, Z. Wang, L. Zhang, L. Zhang W. Liu. J. Shen and D. Tao, Exploring representativeness and informativeness for active learning., IEEE Transactions on Cybernetic, 2017.
[13] Y. Zhang, P. Zhao, J. Cao, W. Ma and J. Huang, Online adaptive asymmetric active learning for budgeted
imbalanced data, SIGKDD International Conference on Knowledge Discovery and Data Mining, 2018.
[14] R. Hu, B. Namee, and S. Delany, Off to a Good Start: Using Clustering to Select the Initial Training Set in
Active Learning., International Florida Artificial Intelligence Research Society Conference (FLAIRS), 2010.
[15] J. Zhu, H. Wang, and B. Tsou, Active learning with sampling by uncertainty and density for word sense disambiguation and text classification., International Conference on Computational Linguistics, 2008.
[16] W. Yuan, Y. Han, D. Guan, S. Lee, and Y.-K. Lee, Initial Training Data Selection for Active Learning.,
International Conference on Ubiquitous Information Management and Communication, 2011.
[17] R. Chattopadhyay, W. Fan, I. Davidson, S. Panchanathan, and J. Ye, Joint transfer and batch-mode active
learning., International Conference on Machine Learning (ICML), 2013.
[18] D. Kale and Y. Liu, Accelerating Active Learning with Transfer Learning., IEEE International Conference on
Data Mining (ICDM), 2013.
[19] Y. Guo, G. Ding, Y. Gao, and J. Han, Active Learning with Cross-Class Similarity Transfer., AAAI Conference
on Artificial Intelligence (AAAI), 2017.
[20] S.-J. Huang and S. Chen, Transfer learning with active queries from source domain., International Joint Conference on Artificial Intelligence (IJCAI), 2016.
[21] Z. Zhou, J. Shin, S. Gurudu, M. Gotway, and J. Liang, AFT*: Integrating Active Learning and Transfer Learning
to Reduce Annotation Efforts., ArXiv preprint, 2018.
[22] Y.-R. Yeh, C.-H. Huang, and Y.-C. Wang, Heterogeneous Domain Adaptation and Classification by Exploiting
the Correlation Subspace., IEEE Transactions on Image Processing, 2014.
[23] Y.-H. Tsai, Y.-R. Yeh, and Y.-C. Wang, Learning Cross-Domain Landmarks for Heterogeneous Domain Adaptation. IEEE International Conference on Computer Vision and Pattern Recognition (CVPR), June. 2016.
[24] C.-A. Hou, Y.-H. Tsai, Y.-R. Yeh and Y.-C. Wang, Unsupervised Domain Adaptation with Label and Structural
Consistency., IEEE Transactions on Image Processing, 2016.
[25] M. Long, J. Wang, G. Ding, J. Sun, and P. S. Yu. Transfer feature learning with joint distribution adaptation.,
IEEE International Conference on Computer Vision, 2013.
[26] S. J. Pan, I. W. Tsang, J. T. Kwok, and Q. Yang. Domain adaptation via transfer component analysis. IEEE
TNN, 2011.
[27] A. Gretton, K. M. Borgwardt, M. J. Rasch, B. Scholkopf, and A. J. Smola. A kernel method for the two-sample
problem. In Proceedings of NIPS, 2006.
[28] B. Settles, Active Learning., Synthesis Lectures on Artificial Intelligence and Machine Learning, 2012.
[29] Z. Wang and J. Ye, Querying Discriminative and Representative Samples for Batch Mode Active Learning.,
ACM Transactions on Knowledge Discovery from Data (TKDD), 2015.
22
[30] S. Chakraborty, V. Balasubramanian, Q. Sun, S. Panchanathan, and J. Ye, Active Batch Selection via Convex
Relaxations with Guaranteed Solution Bounds., IEEE Transactions on Pattern Analysis and Machine Intelligence
(TPAMI), 2015.
[31] S. Kee, E. Castillo, and G. Runger, Query-by-committee improvement with diversity and density in batch active
learning., Information Science, 2018.
[32] W. Cai, M. Zhang, Y. Zhang, Batch Mode Active Learning for Regression With Expected Model Change., IEEE
Transactions on Neural Networks and Learning Systems (TNN), 2017.
[33] I. Lourentzou, D. Gruhl, and S. Welch, Exploring the Efficiency of Batch Active Learning for Human-in-the-Loop
Relation Extraction, The Web Conference (WWW), 2018.
[34] Y. Guo and D. Schuurmans, Discriminative batch mode active learning., International Conference on Neural
Information Processing Systems (NIPS), 2006.
[35] Y. Leng, G-H. Qi, and X.-Y. Xu, A BIC based initial training set selection algorithm for active learning and its
application in audio detection., Radio engineering, 2013.
[36] K. Saenko, B. Kulis, M. Fritz, and T. Darrell, Adapting visual category models to new domains. In European
Conference on Computer Vision (ECCV), 2010.
[37] G. Griffin, A. Holub, and P. Perona, Caltech-256 object category dataset., 2007.
[38] H. Bay, T. Tuytelaars, and L. Van Gool, Surf: Speeded up robust features., In European Conference on Computer
Vision (ECCV). 2006.
[39] B. Gong, Y. Shi, F. Sha, and K. Grauman, Geodesic flow kernel for unsupervised domain adaptation., In
Proceedings of CVPR, 2012.
[40] J. Donahue, Y. Jia, O. Vinyals, J. Hoffman, N. Zhang, E. Tzeng, and T. Darrell, Decaf: A deep convolutional
activation feature for generic visual recognition., In International Conference on Machine Learning (ICML),
2014.
[41] Gong B, Shi Y, Sha F, et al. Geodesic flow kernel for unsupervised domain adaptation[C] Computer Vision and
Pattern Recognition (CVPR), 2012 IEEE Conference on. IEEE, 2012: 2066-2073.