Hello, I'm Shuo Wang, welcome to my page. I'm a dedicated researcher with a focus on the dynamic realm of the financial industry. Additionally, I'm currently pursuing my Data Science master's degree at UC Berkeley through the MIDS program.
My passion lies in the cutting-edge developments within the realms of AI and machine learning. To satisfy my curiosity, I've embarked on several personal projects, allowing me to delve deeper into these fields and stay on top of the latest technological advancements.
Below, you'll find some of my ongoing projects. While they are still in the research stage, my ultimate goal is to refine and develop them into commercially viable products.
I maintain a Medium blog where I share my thoughts on data science and other topics that pique my interest, typically revolving around mathematics or technology.
If you'd like to get in touch with me, the most effective means of communication is through LinkedIn or email. Feel free to reach out!
Action Recognition Utilizing YGAR Data Set
In this project, we developed a simulation system designed to generate diverse yoga action poses. Utilizing these simulated datasets, we conducted training sessions with a range of model architectures to facilitate performance comparisons.
Our primary focus is to harness the capabilities of the simulation system we've created in order to delve into the realm of sequence action recognition, employing the encoder/decoder transformer architecture.
Within the data links, you will find three distinct datasets, each comprising ten unique yoga pose actions. The varying levels of difficulty are attributed to differences in camera angles, zoom distances, and background noise variations.
Natural Language to Data Visualization Using nvBench Data Set
In this project, we embarked on an exploration of the potential applications of large language models (LLMs) and transformer-based models in the realm of natural language translation into data visualization.
Our research leveraged the nvBench dataset, which comprises pairs of natural language queries and corresponding data visualization codes. The ultimate objective was to translate these natural language queries into visualization codes and present them as data visualizations.
For our baseline, we employed the ncNet transformer model from the original research. We did so to investigate the impact of diverse model architectures and the integration of LLMs into the translation process.
Notably, we incorporated the BERT model into a novel transformer architecture to examine the advantages of fine-tuning BERT as the encoder model. Additionally, we explored the potential of CodeT5 by applying this model to the dataset, achieving remarkable results.