service & projects

Summary of my efforts in community service and the connection to my research!

In my time at USC so far, I have been fortunate enough to collaborate with students from the USC Dworak-Peck School of Social Work. I have learned so much from my colleagues and I hope to encourage other students to engage in interdisciplinary work and collaborations as well!

My current experience with community service and NLP

Public opinion expressed on social media can be leveraged as a moving force for policymakers, social movements and other influential political actors in designing effective interventions for homelessness. Due to the nature of the complexities of the homelessness crisis, online discourse on the topic of homelessness elicits a diverse spectrum of attitudes, responses and sentiments that are challenging to capture due to the sheer volume of posts.

Organizations that advocate for equitable housing policies and shelter services face a unique challenge in straddling the interests of community stakeholders and neighborhood residents with those of the unhoused population where online discourse can serve as a valuable indicator for political constituents and advocacy organizations in how to better shape their policy and reform efforts.

Currently, I am an active volunteer with School on Wheels, an organization providing free tutoring and mentoring to children living in shelters, motels, vehicles, group foster homes, and the streets of Southern California. My work with this organization has served as a grounding force for my current research project on Characterizing Attitudes Towards Homelessness.

My past experience with community service and machine learning

As an undergraduate student, my perception of deep learning was blinded by research studies advertising its effectiveness in predictive modeling tasks. However, this changed as I pursued research at the intersection of the non-profit and machine learning domains. I gathered a team of volunteers where our primary goal was to provide quick relief and aid to people experiencing homelessness (PEH) in the DC area but our efforts were not having a lasting impact as PEH migrate to different locations making it difficult to provide aid on a consistent basis. As a student studying machine learning, I became curious about developing ML tools for the non-profit sector.

After participating in the UVA: Save the Children Hackathon. I learned that it is particularly challenging to design models that ensure the privacy of groups at risk. I found it necessary to accompany my research by working with non-profit organizations that could better guide me on the necessary measures for safe and reliable induction of a machine learning system to aid groups in need.

After researching the use of machine learning in the nonprofit space, my findings were sparse and incomplete due to the difficulty of collecting and processing data that is associated with groups of individuals who are inherently difficult to track. As a result, I consulted closely with practitioners from non-profits such as Food for Others, NAMI, Habitat for Humanity and Back on my Feet to better identify the need for such machine learning applications in the non-profit space.

From my research, I found that it would be easier for food banks and shelters to provide aid if they were aware of densely populated areas and migration patterns of groups in need and I started researching the use of machine learning for identifying routes for better resource allocation. I prototyped several architectures that used population density and publicly available records from food distribution centers to identify times and routes for optimal resource allocation. However, after examining existing architectures that addressed a similar problem, I became aware of the societal biases and inequities encoded by these systems and how the lack of data further exacerbates the disparity between user groups. I found that:

  1. The data these systems are reliant on is not fairly sourced, and as a result, leads to an underrepresentation of minority groups
  2. Women specifically are not equally represented and thus, organizations relying on predictive systems grossly underestimate the number of women who are in need of aid. This leads to a shortage of these groups receiving feminine hygiene products.

My work with non-profits has inspired me to pursue research in the topics of fairness and applications of NLP for social good.

Past Projects/Affiliations

  1. Society of PRI: At the University of Virginia, I was an active member of PRI, an organization dedicated to empowering and advocating for minority voices in the CS department.
  2. Society of Women Engineers: Performed service projects at high schools in Charlottesville area to educate minority students about engineering opportunities
  3. Project Clear Skies: UVA HooHacks - Developed a web app using RestAPI that aggregates real time data about a natural disaster from a variety of social media sources giving first responders the ability to perform rapid searches using key words and features. Leveraged Google Vision API and Tensorflow for image classification to provide an accurate assessment of the severity of disasters to reach victims and allocate resources more efficiently.
  4. Save the Children: UVA Data Science Hackathon - Prototyped models in Pytorch for generating infrastructure damage values that can be applied to MDI’s predictive analytics model in an effort to better help with displacement efforts due to disasters.
  5. Truly OpenML: Led a team of four people to pitch a web application that provides a collaborative, intuitive and accessible platform for individuals who are passionate about learning machine learning (ML). Semifinalist at the American Evolution Innovator’s Cup.