Attention Network for Fraud Detection
May 2018 -- Oct 2018
- Designed and implemented a novel neural attention model for attributed sequence classification.
- Integrated conventional sequence attention model with the attributes from user profiles.
- Evaluated the proposed model and compared with state-of-the-art approaches to confirm its effectiveness.
Zhongfang Zhuang, Xiangnan Kong, and Elke Rundensteiner. "AMAS: Attention Model for Attributed Sequence Classification", accepted by SDM 2019. [PDF]
Fraud Detection in One Shot
Dec 2017 -- Apr 2018
- Challenged by the real-world scenario that only one fraud case per fraud type is available.
- Designed a multimodal siamese neural network that is capable of generalizing from only one example.
- Studied and evaluated the proposed model in various real-world scenarios with diverse parameter settings.
Zhongfang Zhuang, Xiangnan Kong, Elke Rundensteiner, Jihane Zouaoui, and Aditya Arora. "One-shot Learning on Attributed Sequences", accepted by IEEE Big Data 2018. [PDF]
Incorporate User Feedback for Fraud Detection
Mar 2017 -- Dec 2017
- Identified the challenges of incorporating the feedback from human domain experts in fraud detection.
- Formulated the problem of deep metric learning on attributed sequences.
- Designed and implemented a deep learning framework to effectively learn from the human feedback.
- Evaluated the purposed model and confirmed it outperforms state-of-the-art in various mining tasks.
Zhongfang Zhuang, Xiangnan Kong, Elke Rundensteiner, Jihane Zouaoui, and Aditya Arora. "Deep Metric Learning on Attributed Sequences", in submission.
Unsupervised Attributed Sequence Embedding
Jan 2016 -- Feb 2017
- Proposed a new data model, the attributed sequence, for Amadeus application log files.
- Identified the challenges of using attributed sequences in fraud detection: attributed sequences are not represented as feature vectors that could be used directly by existing data mining algorithms.
- Designed a multimodal neural network model with a sequence network and an attribute network.
- Tailored an unsupervised training strategy to learn the information from attributed sequences.
- Evaluated the performance of the proposed neural network model in clustering and outlier detection tasks.
- Conducted case studies by using visualization tools and collaborating with domain experts.
Zhongfang Zhuang, Xiangnan Kong, Elke Rundensteiner, Jihane Zouaoui, and Aditya Arora. "Attributed Sequence Embedding", in submission.