Browsing by Author "Li, Yifan"
Now showing 1 - 2 of 2
- Results Per Page
- Sort Options
Item Multistream Classification for Cyber Threat Data with Heterogeneous Feature Space(Association for Computing Machinery, Inc, 2019-05) Li, Yifan; Tao, Hemeng; Gao, Yang; Khan, Latifur; Ayoade, Gbadebo; Thuraisingham, B.; 51656251 (Khan, L); Li, Yifan; Tao, Hemeng; Gao, Yang; Khan, Latifur; Ayoade, Gbadebo; Thuraisingham, B.Under a newly introduced setting of multistream classification, two data streams are involved, which are referred to as source and target streams. The source stream continuously generates data instances from a certain domain with labels, while the target stream does the same task without labels from another domain. Existing approaches assume that domains for both data streams are identical, which is not quite true in real world scenario, since data streams from different sources may contain distinct features. Furthermore, obtaining labels for every instance in a data stream is often expensive and time-consuming. Therefore, it has become an important topic to explore whether labeled instances from other related streams can be helpful to predict those unlabeled instances in a given stream. Note that domains of source and target streams may have distinct features spaces and data distributions. Our objective is to predict class labels of data instances in the target stream by using the classifiers trained by the source stream. We propose a framework of multistream classification by using projected data from a common latent feature space, which is embedded from both source and target domains. This framework is also crucial for enterprise system defenders to detect cross-platform attacks, such as Advanced Persistent Threats (APTs). Empirical evaluation and analysis on both real-world and synthetic datasets are performed to validate the effectiveness of our proposed algorithm, comparing to state-of-the-art techniques. Experimental results show that our approach significantly outperforms other existing approaches. © 2019 IW3C2 (International World Wide Web Conference Committee), published under Creative Commons CC-BY 4.0 License.Item Transfer Learning for Large Scale Data-driven Modeling and Its Applications in Social Science(2021-12-01T06:00:00.000Z) Li, Yifan; Khan, Latifur; Overzet, Lawrence; Brandt, Patrick T.; Ouyang, Jessica J.; Du, Ding-ZhuProliferation of Internet technology in daily life has created numerous data. These data may come from a variety of sources, such as social networks, online businesses, sensors,military surveillance and so on. Generally speaking, these data could be either dynamic or static. Such huge amount of data, along with modern powerful computing capabilities, had pushed forward tremendous amount of applications in AI substantially, including computer vision, natural language processing, cyber-security, and etc. However, most applications still depends heavily on labeled data. The need of massive labeled data hinders the progress of AI research since the labeling process is very laborious and costly, and also goes against how human beings learn - human beings do not require many labeled data by drawing inferences.Transfer learning, as a popular machine learning paradigm recently, leverages knowledge from a source domain to effectively learn predictive models in a target domain which does not have sufficient labeled data. In this dissertation, we propose to investigate transfer learning techniques on an extensive of data-driven applications. First, we unveil a heterogeneous domain adaptation framework on multiple data stream settings, and we apply it to detect cyber attacks within the data stream. Then, we show that multitask learning on multiple domains may actually help with the classification and retrieval tasks on Computer Vision(CV) applications. After the two applications, we shift the research scope to the scenario of offline learning, and test our transfer learning algorithm on sentiment classification task in Natural Language Processing (NLP) domain. We extend the application scenarios even further to the political science domain. We discuss the potential of the modern Graph Neural Networks (GNN) and apply it to the applications of time series forecasting with additional spatial information. Such application includes a number of critical tasks in social science, including but not limited to peace research, transportation analysis, and epidemic spread modeling. We further address modeling challenges observed during the aforementioned applications, such as scalability, model complexity and etc.