Joy He-Yueya

I am a senior studying Computer Science at the University of Washington, where I work on data science and reinforcement learning. I am advised by Tim Althoff. I have also had the fortune of working with Adish Singla at the Max Planck Institute for Software Systems (MPI SWS) and Ying Li at Giving Tech Labs.

Email  /  Twitter

profile photo
News

[12/2020] I received the Computing Research Association (CRA) Outstanding Undergraduate Researcher Award!

Research

My passion lies in using AI to democratize quality education and healthcare. This passion has drawn me toward two main research directions:

  • AI to understand human cognition and behavior: How can we leverage educational and health domain knowledge to build interpretable AI models that can generate actionable insights about human learning and well-being?
  • AI to make decisions: How can we develop AI systems that can guide interventions to improve the learning and health outcomes of individuals?

Modeling and predicitng health conditions

The emergence of mobile devices that can passively collect detailed behavioral data, such as smartphones and smartwatches, provides us with an unprecedented opportunity to develop computational methods that can generate and apply insights about human well-being at scale.

In one project, I studied the relationship between behavioral stability and schizophrenia symptoms, leveraging data generated by a passive sensing system carried by over 60 individuals with schizophrenia for one year. I developed a novel metric that quantifies the stability of various passively sensed behaviors and is predictive of psychiatric symptom severity.

Assessing the relationship between routine and schizophrenia symptoms with passively sensed measures of behavioral stability
Joy He-Yueya, Benjamin Buck, Andrew Campbell, Tanzeem Choudhury, John M. Kane, Dror Ben-Zeev, Tim Althoff
npj schizophrenia, 2020. [pdf]

Generating personalized curricula

I'm interested in exploring how AI can support educators through automation of curriculum design. Reinforcement learning (RL) is a powerful framework for learning a decision policy through experience to optimize future outcomes. I've been working on developing RL methods for learning tasks recommedation (i.e., determining a personalized sequence of questions for each student based on their previous interactions with existing exercises to best help them learn all the relevant concepts).


Student knowledge diagnosis

AI could also help educators learn the student's knowledge state. The lack of interpretability and generalizability of state-of-the-art knowledge tracing models based on neural networks limits their ability to provide educators with actionable insights about how students are learning, which concepts they are struggling with, and what misconceptions they have. I am excited about overcoming these limitations by applying theoretical frameworks in cognitive science to AI models.


Voice coaching for quality teaching

To help educators and caregivers speak more appropriately to their audience, my team at Giving Tech Labs and I built a real-time automatic voice coaching app that analyzes various acoustis features that can be mapped to key auditory measures such as speaking rate, energy level, and intonation. This app offers a promising tool for educators to modulate their voice and improve their interactions with students.

Miscellany

I'm always happy to help undergraduate students get involved in research. Feel free to email me!

Opportunities related to AI for education: