State-Relabeling Adversarial Active Learning

CVPR, pp. 8753-8762, 2020.

Cited by: 0|Bibtex|Views60|Links
EI
Keywords:
deep neural networklearning methodunified representationdeep learningprevious stateMore(17+)
Weibo:
The experiments on image classification and segmentation demonstrate that our model outperforms previous state-of-the-art methods and the initially sampling algorithm significantly improve the performance of our model

Abstract:

Active learning is to design label-efficient algorithms by sampling the most representative samples to be labeled by an oracle. In this paper, we propose a state relabeling adversarial active learning model (SRAAL), that leverages both the annotation and the labeled/unlabeled state information for deriving the most informative unlabeled...More

Code:

Data:

0
Introduction
  • Deep neural network models have made great success in many areas, they still heavily rely on large-scale labeled data to train large number of parameters
  • It is very difficult, time-consuming, or expensive to obtain labeled samples, which becomes the main bottleneck for deep learning methods [10].
  • As shown in Fig. 1, active learning algorithm is typically an iterative process in which a set of samples is selected to be labeled from an unlabeled pool at each iteration.
  • How to select the most informative samples from the unlabeled pool is the key problem in active learning
Highlights
  • Deep neural network models have made great success in many areas, they still heavily rely on large-scale labeled data to train large number of parameters
  • We propose a state relabeling adversarial active learning model (SRAAL) that considers both the annotation and the state information for deriving most informative unlabeled samples
  • To better use the state information, we propose the online uncertainty indicator (OUI) to calculate an uncertainty score to relabel the state of unlabeled data
  • We study the active learning and propose a state-relabeling adversarial active learning model (SRAAL) that makes full use of both the annotation and the state information for deriving most informative unlabeled samples
  • The model consists of a unified representation generator that learns the annotation-embedded image feature, and a labeled/unlabeled state discriminator that selects most informative samples with the help of online updated indicator
  • The experiments on image classification and segmentation demonstrate that our model outperforms previous state-of-the-art methods and the initially sampling algorithm significantly improve the performance of our model
Methods
  • There exists a large unlabeled data pool from which the authors can randomly select M samples and obtain annotations of them via an oracle.
  • Let them denote the initial unlabeled pool by DU and the initial labeled pool by DL.
  • Denotes that a data point in unlabeled pool and denotes a data point and its annotation in labeled pool.
  • The key of the AL algorithm is to select the most informative samples from the unlabeled pool DU.
  • As illustrated in Fig. 1, this procedure will repeat until the performance of the target model meets user’s requirements, or the budget for annotation runs out
Results
  • Experiments on four datasets at image classification and segmentation tasks show that the method outperforms previous state-of-the-art methods.
  • The highest accuracy of the Resnet-18 with full dataset reaches 93.5%, which is only 1.02% better than SRAAL with 40% samples.
  • The authors' SRAAL achieves the better performance than the state-of-the-art methods, such as VAAL, LL4AL.
  • The experiments on image classification and segmentation demonstrate that the model outperforms previous state-of-the-art methods and the initially sampling algorithm significantly improve the performance of the model
Conclusion
  • The authors study the active learning and propose a state-relabeling adversarial active learning model (SRAAL) that makes full use of both the annotation and the state information for deriving most informative unlabeled samples.
  • The authors introduce the k-center approach to initialize the labeled pool, which makes subsequent sampling more efficient.
  • The experiments on image classification and segmentation demonstrate that the model outperforms previous state-of-the-art methods and the initially sampling algorithm significantly improve the performance of the model
Summary
  • Introduction:

    Deep neural network models have made great success in many areas, they still heavily rely on large-scale labeled data to train large number of parameters
  • It is very difficult, time-consuming, or expensive to obtain labeled samples, which becomes the main bottleneck for deep learning methods [10].
  • As shown in Fig. 1, active learning algorithm is typically an iterative process in which a set of samples is selected to be labeled from an unlabeled pool at each iteration.
  • How to select the most informative samples from the unlabeled pool is the key problem in active learning
  • Methods:

    There exists a large unlabeled data pool from which the authors can randomly select M samples and obtain annotations of them via an oracle.
  • Let them denote the initial unlabeled pool by DU and the initial labeled pool by DL.
  • Denotes that a data point in unlabeled pool and denotes a data point and its annotation in labeled pool.
  • The key of the AL algorithm is to select the most informative samples from the unlabeled pool DU.
  • As illustrated in Fig. 1, this procedure will repeat until the performance of the target model meets user’s requirements, or the budget for annotation runs out
  • Results:

    Experiments on four datasets at image classification and segmentation tasks show that the method outperforms previous state-of-the-art methods.
  • The highest accuracy of the Resnet-18 with full dataset reaches 93.5%, which is only 1.02% better than SRAAL with 40% samples.
  • The authors' SRAAL achieves the better performance than the state-of-the-art methods, such as VAAL, LL4AL.
  • The experiments on image classification and segmentation demonstrate that the model outperforms previous state-of-the-art methods and the initially sampling algorithm significantly improve the performance of the model
  • Conclusion:

    The authors study the active learning and propose a state-relabeling adversarial active learning model (SRAAL) that makes full use of both the annotation and the state information for deriving most informative unlabeled samples.
  • The authors introduce the k-center approach to initialize the labeled pool, which makes subsequent sampling more efficient.
  • The experiments on image classification and segmentation demonstrate that the model outperforms previous state-of-the-art methods and the initially sampling algorithm significantly improve the performance of the model
Tables
  • Table1: Comparison with entropy and standard deviation(SD) under different sampling ratios
Download tables as Excel
Related work
  • Active learning has been widely studied for decades and most of the classical methods can be grouped into three scenarios: membership query synthesis [1], stream-based selective sampling [5, 23] and pool-based sampling. As the acquirement of abundant unlabeled samples becomes easy, most of recent works [14, 37, 2, 7, 39] focus on the last scenarios. Current active learning methods can be divided into two categories: pool-based approaches and synthesizing approaches.

    Instead of querying most informative instances from an unlabeled pool, the synthesizing approaches [31, 32, 47] use generative models to produce new synthetic samples that are informative for the current model. These methods typically introduce various GAN models [16, 33] or VAE models [22, 40] into their algorithm to generate informative data with high quality. However, the synthesizing approaches still has some disadvantages to overcome, such as high computational complexity and instability of performance [47]. For this reason, this paper mainly focuses on research of the pool-based approaches.
Funding
  • This work was supported in part by the National Key R&D Program of China under Grand:2018AAA0102003, in part by National Natural Science Foundation of China: 61771457, 61732007, 61772497, 61772494, 61931008, 61620106009, U1636214, 61622211, U19B2038, and in part by Key Research Program of Frontier Sciences, CAS: QYZDJSSW-SYS013 and the Fundamental Research Funds for the Central Universities under Grant WK2100100030
Reference
  • Dana Angluin. Queries and concept learning. Machine learning, 2(4):319–342, 1988.
    Google ScholarLocate open access versionFindings
  • William H Beluch, Tim Genewein, Andreas Nurnberger, and Jan M Kohler. The power of ensembles for active learning in image classification. In IEEE Conference on Computer Vision and Pattern Recognition, pages 9368–9377, 2018.
    Google ScholarLocate open access versionFindings
  • Mustafa Bilgic and Lise Getoor. Link-based active learning. In NIPS Workshop on Analyzing Networks and Learning with Graphs, 2009.
    Google ScholarFindings
  • Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. The cityscapes dataset for semantic urban scene understanding. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 3213–3223, 2016.
    Google ScholarLocate open access versionFindings
  • Ido Dagan and Sean P Engelson. Committee-based sampling for training probabilistic classifiers. In Machine Learning Proceedings 1995, pages 150–157.
    Google ScholarLocate open access versionFindings
  • Carl Doersch, Abhinav Gupta, and Alexei A Efros. Unsupervised visual representation learning by context prediction. In Proceedings of the IEEE International Conference on Computer Vision, pages 1422–1430, 2015.
    Google ScholarLocate open access versionFindings
  • In So Kweon Donggeun Yoo. Learning loss for active learning. Technical report, University of Wisconsin-Madison Department of Computer Sciences, 2019.
    Google ScholarFindings
  • Melanie Ducoffe and Frederic Precioso. Adversarial active learning for deep networks: a margin based approach. arXiv preprint arXiv:1802.09841, 2018.
    Findings
  • Sayna Ebrahimi, Mohamed Elhoseiny, Trevor Darrell, and Marcus Rohrbach. Uncertainty-guided continual learning with bayesian neural networks. arXiv preprint arXiv:1906.02425, 2019.
    Findings
  • Sayna Ebrahimi, Anna Rohrbach, and Trevor Darrell. Gradient-free policy architecture search and adaptation. arXiv preprint arXiv:1710.05958, 2017.
    Findings
  • Ehsan Elhamifar, Guillermo Sapiro, Allen Yang, and S Shankar Sasrty. A convex optimization framework for active learning. In Proceedings of the IEEE International Conference on Computer Vision, pages 209–216, 2013.
    Google ScholarLocate open access versionFindings
  • Li Fei-Fei, Rob Fergus, and Pietro Perona. Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories. Computer Vision and Pattern Recognition Workshop, 2004.
    Google ScholarLocate open access versionFindings
  • Alexander Freytag, Erik Rodner, and Joachim Denzler. Selecting influential examples: Active learning with expected model output changes. In European Conference on Computer Vision, pages 562–577.
    Google ScholarLocate open access versionFindings
  • Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. In international conference on machine learning, pages 1050–1059, 2016.
    Google ScholarLocate open access versionFindings
  • Teofilo F. Gonzalez. Clustering to minimize the maximum intercluster distance. Theoretical Computer Science, 38(none):293–306.
    Google ScholarLocate open access versionFindings
  • Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In NIPS, pages 2672–2680, 2014.
    Google ScholarLocate open access versionFindings
  • Marc Gorriz, Axel Carlier, Emmanuel Faure, and Xavier Giro-i Nieto. Cost-effective active learning for melanoma segmentation. arXiv preprint arXiv:1711.09168, 2017.
    Findings
  • Mahmudul Hasan and Amit K Roy-Chowdhury. Context aware active learning of activity recognition models. In Proceedings of the IEEE International Conference on Computer Vision, pages 4543–4551, 2015.
    Google ScholarLocate open access versionFindings
  • Armand Joulin, Laurens van der Maaten, Allan Jabri, and Nicolas Vasilache. Learning visual features from large weakly supervised data. In European Conference on Computer Vision, pages 67–84.
    Google ScholarLocate open access versionFindings
  • Christoph Kading, Erik Rodner, Alexander Freytag, and Joachim Denzler. Active and continuous exploration with deep neural networks and expected model output changes. arXiv preprint arXiv:1612.06129, 2016.
    Findings
  • Ashish Kapoor, Kristen Grauman, Raquel Urtasun, and Trevor Darrell. Active learning with gaussian processes for object categorization. In 2007 IEEE 11th International Conference on Computer Vision, pages 1–8. IEEE, 2007.
    Google ScholarLocate open access versionFindings
  • Diederik P Kingma and Max Welling. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013.
    Findings
  • Vikram Krishnamurthy. Algorithms for optimal scheduling and management of hidden markov model sensors. IEEE Transactions on Signal Processing, 50(6):1382–1397, 2002.
    Google ScholarLocate open access versionFindings
  • Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. Technical report, Citeseer, 2009.
    Google ScholarFindings
  • Weicheng Kuo, Christian Hne, Esther Yuh, Pratik Mukherjee, and Jitendra Malik. Cost-sensitive active learning for intracranial hemorrhage detection.
    Google ScholarFindings
  • Liang Li, Shuqiang Jiang, and Qingming Huang. Learning hierarchical semantic description via mixed-norm regularization for image understanding. IEEE Transactions on Multimedia, 14(5):1401–1413, 2012.
    Google ScholarLocate open access versionFindings
  • Xin Li and Yuhong Guo. Adaptive active learning for image classification. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 859–866, 2013.
    Google ScholarLocate open access versionFindings
  • Xuejing Liu, Liang Li, Shuhui Wang, Zheng-Jun Zha, Dechao Meng, and Qingming Huang. Adaptive reconstruction network for weakly supervised referring expression grounding. In IEEE/CVF International Conference on Computer Vision, ICCV, pages 2611–2620. IEEE, 2019.
    Google ScholarLocate open access versionFindings
  • Oisin Mac Aodha, Neill DF Campbell, Jan Kautz, and Gabriel J Brostow. Hierarchical subquery evaluation for active learning on a graph. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 564–571, 2014.
    Google ScholarLocate open access versionFindings
  • Dhruv Mahajan, Ross Girshick, Vignesh Ramanathan, Kaiming He, Manohar Paluri, Yixuan Li, Ashwin Bharambe, and Laurens van der Maaten. Exploring the limits of weakly supervised pretraining. In Proceedings of the European Conference on Computer Vision (ECCV), pages 181–196, 2018.
    Google ScholarLocate open access versionFindings
  • Dwarikanath Mahapatra, Behzad Bozorgtabar, Jean-Philippe Thiran, and Mauricio Reyes. Efficient active learning for image classification and segmentation using a sample selection and conditional generative adversarial network. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pages 580–588.
    Google ScholarLocate open access versionFindings
  • Christoph Mayer and Radu Timofte. Adversarial sampling for active learning. arXiv preprint arXiv:1808.06671, 2018.
    Findings
  • Mehdi Mirza and Simon Osindero. Conditional generative adversarial nets. arXiv preprint arXiv:1411.1784, 2014.
    Findings
  • Hieu T Nguyen and Arnold Smeulders. Active learning using pre-clustering. In ICML, page 79. ACM, 2004.
    Google ScholarLocate open access versionFindings
  • Mehdi Noroozi, Hamed Pirsiavash, and Paolo Favaro. Representation learning by learning to count. In Proceedings of the IEEE International Conference on Computer Vision, pages 5898–5906, 2017.
    Google ScholarLocate open access versionFindings
  • Nicholas Roy and Andrew McCallum. Toward optimal active learning through monte carlo estimation of error reduction. ICML, Williamstown, pages 441–448, 2001.
    Google ScholarFindings
  • Ozan Sener and Silvio Savarese. Active learning for convolutional neural networks: A core-set approach. arXiv preprint arXiv:1708.00489, 2017.
    Findings
  • Burr Settles, Mark Craven, and Soumya Ray. Multipleinstance active learning. In Advances in neural information processing systems, pages 1289–1296, 2008.
    Google ScholarLocate open access versionFindings
  • Samarth Sinha, Sayna Ebrahimi, and Trevor Darrell. Variational adversarial active learning. arXiv preprint arXiv:1904.00370, 2019.
    Findings
  • Kihyuk Sohn, Honglak Lee, and Xinchen Yan. Learning structured output representation using deep conditional generative models. In Advances in neural information processing systems, pages 3483–3491, 2015.
    Google ScholarLocate open access versionFindings
  • Ke Sun, Bin Xiao, Dong Liu, and Jingdong Wang. Deep high-resolution representation learning for human pose estimation. In IEEE Conference on Computer Vision and Pattern Recognition, CVPR, pages 5693–5703, 2019.
    Google ScholarLocate open access versionFindings
  • Lin Yang, Yizhe Zhang, Jianxu Chen, Siyuan Zhang, and Danny Z Chen. Suggestive annotation: A deep active learning framework for biomedical image segmentation. In International conference on medical image computing and computer-assisted intervention, pages 399–407.
    Google ScholarLocate open access versionFindings
  • Shijie Yang, Liang Li, Shuhui Wang, Weigang Zhang, Qingming Huang, and Qi Tian. Skeletonnet: A hybrid network with a skeleton-embedding process for multi-view image representation learning. IEEE Transactions on Multimedia, 21(11):2916–2929, 2019.
    Google ScholarLocate open access versionFindings
  • Yi Yang, Zhigang Ma, Feiping Nie, Xiaojun Chang, and Alexander G Hauptmann. Multi-class active learning by uncertainty sampling with diversity maximization. International Journal of Computer Vision, 113(2):113–127, 2015.
    Google ScholarLocate open access versionFindings
  • Zheng-Jun Zha, Meng Wang, Yan-Tao Zheng, Yi Yang, Richang Hong, and Tat-Seng Chua. Interactive video indexing with statistical active learning. IEEE Trans. Multimedia, 14(1):17–27, 2012.
    Google ScholarLocate open access versionFindings
  • Zhi Hua Zhou. A brief introduction to weakly supervised learning. National Science Review, v.5(1):48–57.
    Google ScholarLocate open access versionFindings
  • Jia-Jie Zhu and Jose Bento. Generative adversarial active learning. arXiv preprint arXiv:1702.07956, 2017.
    Findings
Your rating :
0

 

Tags
Comments