Zhichen Gong, IEEE Student Member

Mater of Science

Rm 606, West Building of Science and Technoly
School of Computer Science and Technology
University of Science and Technology of China
Hefei, Anhui, China (安徽省合肥市中国科学技术大学计算机学院)

Email: zcgong@mail.ustc.edu.cn

Wechat: zcgustc

Biography [CV]

Hi, I am currently a third year Master student in the Department of Computer Science, The University of Science and Technology of China, supervised by Prof. Huanhuan Chen. I am with the USTC-Birmingham Joint Research Institute(UBRI) since 2014. I received the B. Eng degree from Department of Computer Science and Technology in Anhui University in 2015.

My research interest includes machine learning, Reinforcement learning, neural dynamic learning systems and deep generative models. I am also interested in NLP.

I am a researcher and developer in machine learning, the art in many domains of which is really fascinating to me. Send me an email if you would like to get in touch, have questions or feedback on my work, or just want to say hello.

News

[09/2016] The paper on kernel learning for Imbalanced data oversampling is accepted by CIKM.

Selected Publications | My Google Scholar

	Zhichen Gong, Huanhuan Chen, Bo Yuan, and Xin Yao. "Multi-objective Learning in the Model Space for Time Series Classification" IEEE Trans. on Cybernetics (Accepted), 2017. [paper] [code]
	Zhichen Gong, Huanhuan Chen. "Model-Based Oversampling for Imbalanced Sequence Classification" ACM International Conference on Information and Knowledge Management (CIKM), 2016. (full Paper) [paper] The imbalance of data distribution has been shown to have an adverse inﬂuence on the recall of the minority class. A typical solution is oversampling the minority class, which, however, may be invalidated by class overlapping and nonlinearity, resulting in overﬁtting. My idea is to oversample the minority class in the model-kernel feature space, where the data are more likely to be linearly separated so that the sampling could be safer. We perform experiments on imbalanced univariate and multivariate time series datasets. It achieves better robustness and signiﬁcantly better performance on F1 measure, G-means and AUC on most datasets.
	Zhichen Gong, Huanhuan Chen. "Sequential Data Classification by Dynamic State Warping" Journal of Knowledge and Information System (Accepted), 2017. (-) [paper] [code] A very successful tempral alignment algorithm, Dynamic time warping (DTW) is to compare frammes in the observation space for alignment and returns a distance value. However, the point-level comparison makes it have weak interpretation and sensitive to noise. We propose to improve DTW by considering the dynamical information with a nonlinear yet efficient dynamical system. The alignment is based on the comparison of states that are aware of the input history, intead of raw frames. The experiments are performed on all 86 UCR datasets to evaluate the classification performance, robustness, scalability, and parameter sensitivity.
	Zhichen Gong, Huanhuan Chen. "Model Space Learning by Point-Level Metric for Sequence Classification" manuscript in submission, 2017. [paper] [Code]
	Zhengfu Liu, Zhenyu Liu, Yinlong Song,Zhichen Gong, Huanhuan Chen. "Predicting Stock Trend Using Multi-objective Diversified Echo State Network" IEEEInternational Conference onInformation Science and Technology. (ICIST), 2017. (As the idea contributor) [paper] [code]
	Dandan Yang, Huanhuan Chen, Yinlong Song, and Zhichen Gong. "Granger Causality for Multivariate Time Series Classification" IEEE International Conference on Big Knowledge (ICBK), 2017. (As the idea contributor) [paper] [code]
	Zhichen Gong, Yan Liu, Huanhuan Chen. "Learning Dynamic Spatio-Temporal Relations for Human Activity Recognition" manuscript in submission (-), 2017. [paper] [code]