9 March 2022 in Monthly Notices of the Royal Astronomical Society
Pulsar observation and research are of great significance. With the gradual increase in the performance and quantity of observing equipment, the received pulsar observation data also increase geometrically. Machine learning can mine large-scale pulsar data sets to realize automatic pulsar candidate identification. However, the pulsar candidate and non-pulsar candidate data sets are highly imbalanced. The traditional classifier trained on the data set with imbalanced class distribution usually shows poor generalization performance, which will cause the classifier to be biased towards negative instances. To address the problem of poor identification algorithm performance caused by sample imbalance, we describe a new framework, namely AdaBoost-multi-input-CNN (AdaBoost-MICNN). In AdaBoost, multiple baseline models are trained sequentially, and the weight of each training sample changes as the training progresses. Especially the minority samples, which are more easily ignored, will get more attention in subsequent baseline models. This will solve the problem of imbalanced classification. Meanwhile, there are usually four pulsar diagnostic plots. To automatically extract and integrate these four diagnostic plots features, we use multi-input-CNN (MICNN) as the baseline model of AdaBoost. The attention mechanism is introduced to enhance the performance of MICNN’s extraction of features, and design an objective function that adds the maximum mean discrepancy to enhance the anti-interference ability of the model. Finally, the proposed method is tested in the High Time Resolution Universe Medlat Data, and the experimental results verify the effectiveness and efficiency of the method.
17 November 2022 in Journal of Obstetrics and Gynaecology Research
24 February 2022 in arXiv
Machine Reading Comprehension(MRC) has achieved a remarkable result since some powerful models, such as BERT, are proposed. However, these models are not robust enough and vulnerable to adversarial input perturbation and generalization examples. Some works tried to improve the performance on specific types of data by adding some related examples into training data while it leads to degradation on the original dataset, because the shift of data distribution makes the answer ranking based on the softmax probability of model unreliable. In this paper, we propose a method to improve the robustness by using a calibrator as the post-hoc reranker, which is implemented based on XGBoost model. The calibrator combines both manual features and representation learning features to rerank candidate results. Experimental results on adversarial datasets show that our model can achieve performance improvement by more than 10\% and also make improvement on the original and generalization datasets.