Honggen Zhang

I am a fourth-year Ph.D. candidate majoring in Electrical and Computer Engineering at the University of Hawaii at Manoa. Currently, I am working with Dr.June Zhang, sponsored by the AI Institute in Dynamic Systems. I got my M.S. in mathematical at North China University of Technology in 2019.

Research Interest and Skills

I have a broad interest in representation learning, machine learning, and self-supervised learning. I am correctly working on tabular data representation learning and Dynamic system representation learning.

• Dynamic system embedding.
• Machie learning on Health dataset
• Self-supervised learning.
• Statistic learning.

A Few Publications

The following is my current research work.

Knowledge graph embedding [Arxiv preprint]

I am currently working on developing a machine-learning model that utilizes non-labeled data. My research is focused on knowledge graphs, which are comprised of triples containing factual information about the world (head entity, relationship, tail entity). However, due to the incompleteness of knowledge graphs, I am pursuing the task of knowledge graph completion. To address this problem, I approach it as a fill-in-the-blank challenge, where missing parts of the knowledge graph are predicted using the available data. To improve the accuracy of these predictions, have proposed a debiased contrastive knowledge graph embedding technique. This involves generating non-factual triples to contrast with the factual ones, resulting in a more accurate model. I am proud to say that our method has achieved state-of-the-art results in knowledge graph completion.

How does news evolve? [Arxiv preprint]

The spread of fake rumors is a major problem that can have a significant impact on the dissemination of true news. One of the key differences between fake rumors and true news is the writing style, which can evolve over time. Therefore, our objective is to identify these differences by scraping news articles from various websites on a daily and collecting a large dataset. Our goal is to model how a particular event evolves over time through different writing styles. To accomplish this, we are leveraging information extraction (IE) methods to extract object- relationship-object triplets from the text. We will then use natural language processing tools to transform the unstructured data into a structured dataset that we can analyze.

ECG data with ML

We are collaborating with our local hospital to enhance the accuracy of disease diagnosis using ECG data. Our goal is to develop a machine learning model that can classify patients with and without a disease based on their ECG data. However, dealing with an unbalanced dataset has been challenging, as the number of patients with the disease is significantly less than those without it.To address this issue, we have recently augmented the disease data to balance it with the healthy data, which has resulted in improved classification results. We are excited about the potential of our approach to enhance disease diagnosis accuracy and improve patient outcomes.

Honors