Responsibilities include:

• Analyzed the unique property of graph fairness data.
• Aimed to boost the performance of GNN model on fairness graph data while alleviating the oversmoothing problem.
• Enriched the representation diversity of model from three perspectives: the node embedding of fairness graph data, learned representation across different experts in Mixed of Expert (MOE), and learned representation across different-depth layers.
• Analyzed loss landscape of the model to empirically and theoretically explain the effectiveness of diversity learning.
• Successfully improved the accuracy of a newly designed GNN based model under adversarial training and fairness data while restraining the oversmooth problem from happening.

Responsibilities include:

• Signified harmful posts/comments manually from more than 60k tweets to serve as positive/negative samples during the training process.
• Collected project related information (e.g. BERT, GNN and contrasive learning), and helped run experiments on baseline models.
• Helped run the experiments on new designed model over two tasks: malicious user detection and malicious tweet detection.
• Assisted the first author to draft a manuscript of the work and submitted to ICDM conference.

Responsibilities include:

• Designed the website through front end languages (HTML, CSS, Javascript, AJAX, etc.) to display the relationship between the protein groups of different types of coronavirus with different hosts.
• Imported the previous work data of the paper from github to construct website, which can be used to identify potential infected host groups based on similar structure of the protein groups. (Simple version of the website)
• Updated website based on the most recent result given by the algorithm.

Responsibilities include:

• Designed a website to hold data of a machine learning algorithm that predicted the COVID-19 trend based on the ground truth using front end languages (Javascript, HTML, CSS, etc. ) and python.
• Pulled data from Airtable and transferred it using Python to make it compatible with the website.
• Modified the website to be user-friendly, which also serves as a platform to compare the groundtruth/predicted cases of each state.
• Co-published the work to medRxiv.