Batch mode active learning via adaptive criteria weights
- PDF / 1,167,175 Bytes
- 15 Pages / 595.224 x 790.955 pts Page_size
- 88 Downloads / 219 Views
Batch mode active learning via adaptive criteria weights Hao Li1 · Yongli Wang2
· Yanchao Li1 · Gang Xiao3 · Peng Hu1 · Ruxin Zhao1
Accepted: 16 September 2020 © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Batch mode active learning (BMAL) is absorbed in training reliable classifier with deficient labeled examples by efficiently querying the most valuable unlabeled examples for supervision. In particular, BMAL always selects examples based on the decent-designed criteria, such as (un)certainty and representativeness, etc. However, existing BMAL approaches make a naive trade-off between the criteria and simply combine them with fixed weights, which may yield suboptimal batch selection since the criteria of unlabeled examples would fluctuate after retraining classifier with the newly augmented training set as the learning of classifier progresses. Instead, the weights of the criteria should be assigned properly. To overcome this problem, this paper proposes a novel Adaptive Criteria Weights active learning method, abbreviated ACW, which dynamically combines the example selection criteria together to select critical examples for semi-supervised classification. Concretely, we first assign an initial value to each criterion weight, then the current optimal batch is picked from unlabeled pool. Thereafter, the criteria weights are learned and adjusted adaptively by minimizing the objective function with the selected batch at each round. To the best of our knowledge, this work is the first attempt to explore adaptive criteria weights in the context of active learning. The superiority of ACW against the existing state-of-the-art BMAL approaches has also been validated by extensive experimental results on widely used datasets. Keywords Batch mode active learning · Adaptive criteria weights · Classification
1 Introduction There are abundant unlabeled examples and scarce labeled examples lying in many data mining, pattern recognition and natural language processing tasks [1, 37, 44, 56], where training reliable prediction model is a challenging issue since manually labeling clean training set is extremely labor-intensive, time-consuming and error-prone [19, 36, 43]. One vivid example is computer-aided medical diagnosis, it is easily to obtain large numbers, of chest XRay images from medical examinations, but determining Yongli Wang
[email protected] Hao Li [email protected] 1
School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, China
2
School of Computer Science, Nanjing University of Posts and Telecommunications, Nanjing, China
3
Science and technology on Complex Systems Simulation Laboratory, Nanjing, China
which disease a specified X-Ray image belongs to for a physician is impractical. Accordingly, active learning [20, 31, 33, 39, 45, 61] is introduced to improve the generalization ability of learning model while saving labeling efforts. The key idea behind active learning is that the algorithm selects the most “valuable” ex
Data Loading...
