Liyan Song
Research Assistant Professor
Positions :
College of Engineering Department of Computer Science and Engineering
Research Contents : My main research area focuses on machine learning and its application to predictive software engineering (i.e., software defect prediction and software effort estimation).
My (the other) main research interest is the classical machine learning (ML), including regression and classification, statistical ML, Bayesian regression, class imbalance learning, online learning, ensemble learning and so on. I myself also had done a few works on unsupervised ML, specifically independent component analysis (ICA), in the gap year of my PhD period.
Together with some PhD students in our group, I also work with deep neural network (DNN) on image data. I know the basic things, being capable of making analysis on the networks, but I am not good at constructing / designing the architecture of deep networks though people are always telling me that to be simple, neither do I be familiar with those fancy, tricky and interesting skills to make DNNs working well.
Regarding the application of software engineering, I am working with the below two tasks:
1) Justintime software defect prediction (JITSDP) is typically a binary classification problem with highly class imbalance. Machine learning skills required include binary classification, online learning, imbalance learning, dealing with label noise and so on. We can explore and dig out experimental data from GitHub.
2) Software effort estimation (SEE) is a regression problem with small data and highly "label noise". The datasets are scare and the data in each dataset is small. The data quality is NOT high.
ORCID : 0000000311728825
Scopus Author ID : 56313668000
Contact email :
songly@sustech.edu.cn
BiographyMy main research area focuses on machine learning and its application to predictive software engineering (i.e., software defect prediction and software effort estimation). My (the other) main research interest is the classical machine learning (ML), including regression and classification, statistical ML, Bayesian regression, class imbalance learning, online learning, ensemble learning and so on. I myself also had done a few works on unsupervised ML, specifically independent component analysis (ICA), in the gap year of my PhD period. Together with some PhD students in our group, I also work with deep neural network (DNN) on image data. I know the basic things, being capable of making analysis on the networks, but I am not good at constructing / designing the architecture of deep networks though people are always telling me that to be simple, neither do I be familiar with those fancy, tricky and interesting skills to make DNNs working well. Regarding the application of software engineering, I am working with the below two tasks: 1) Justintime software defect prediction (JITSDP) is typically a binary classification problem with highly class imbalance. Machine learning skills required include binary classification, online learning, imbalance learning, dealing with label noise and so on. We can explore and dig out experimental data from GitHub. 2) Software effort estimation (SEE) is a regression problem with small data and highly "label noise". The datasets are scare and the data in each dataset is small. The data quality is NOT high. Employment
Education
