Kun Bu, a postdoctoral researcher at the University of South Florida, is building statistical and machine learning methods to extract reliable insights from massive, interconnected datasets. Her work addresses a core challenge in modern science: how to turn raw data into trustworthy knowledge that can guide decisions in healthcare, finance, and public policy.
Bu's path to data science began with an early exposure to finance, where she saw "how decisions affecting millions of people were often made under uncertainty," she said. That experience drove her to understand how data can inform more reliable choices. She found that the same statistical principles apply across disciplines, from studying medication safety to modeling human behavior, and she was drawn to the field's interdisciplinary nature.
Advancing data integrity with machine learning
Kun Bu's research develops new statistical and AI approaches that can identify meaningful patterns, quantify uncertainty, and support better decision-making. Many critical problems in healthcare, finance, and public policy generate enormous amounts of information, but traditional statistical methods often struggle to capture hidden relationships within these datasets. "I was especially attracted to the interdisciplinary nature of the field, where mathematics, computing, and domain expertise come together to solve real-world problems," she said.
For example, Bu has worked on methods to detect medication safety risks using healthcare databases. A common theme across her projects is building tools that help researchers move beyond simple prediction toward deeper understanding of underlying systems. Ultimately, her goal is to make data-driven discoveries more accurate, interpretable, and useful for scientists, clinicians, and policymakers.
Overcoming misconceptions in AI
One of the most surprising lessons from working with large datasets, Bu said, is that "having more data does not automatically lead to better understanding." Early in her career, she assumed larger datasets would produce clearer answers. Instead, she found that big data often introduces hidden biases, missing information, and complex relationships that can mislead analysts if not carefully examined.
She also learned that sophisticated AI and statistical models still leave significant uncertainty. "The most valuable insights often come not from finding a single 'correct' answer, but from understanding the range of possible explanations and quantifying confidence in the results," Bu said. This insight shifted her focus toward interpretability, transparency, and uncertainty quantification, rather than just building more powerful predictive models.
Combining AI with statistical reasoning
Bu is now most excited about developing methods that combine AI's pattern recognition with statistics' strength in interpretation and uncertainty. "I am particularly interested in developing methods that integrate the predictive power of AI with the interpretability and uncertainty quantification of statistics," she said. Many AI models still operate as "black boxes," she explained, making it difficult to trust their predictions. By revealing the underlying relationships that drive patterns, these tools could help decision-makers move from prediction to genuine understanding in healthcare, finance, and other scientific domains.
Her hope is that this work contributes to a future where data are used not simply to automate decisions, but to enhance human understanding and support more thoughtful, evidence-based choices in science, industry, and society.
Why this matters for science and research
For scientists and researchers, Bu's work underscores a critical shift in data analysis: predictive accuracy alone is not enough. As datasets grow, the risk of hidden biases and spurious correlations increases. Her focus on interpretability and uncertainty quantification provides a roadmap for building tools that not only find patterns but also explain them-essential for high-stakes applications like drug safety surveillance and public health policy. The push for AI for Science & Research that is both accurate and interpretable is gaining momentum, and Bu's methods offer a concrete path forward. Researchers who adopt these approaches can produce findings that are more transparent, reproducible, and actionable.
Your membership also unlocks:
