AI/Machine Learning in Healthcare
The Healthcare domain is uniquely rewarding and challenging. The decisions taken on the basis of machine learning and AI systems can greatly improve or worsen a person’s quality of life. My research is focused on risk prediction at various stages of patient’s life cycle e.g., disease progression, risk of mortality, algorithms for care plan management etc. I work at KenSci which is an AI in healthcare focused startup, spun off from University of Washington.
Accountability and Explainability in AI
Given the increasing proliferation of AI system, there is a need to hold such systems accountable as their decisions can have severe consequence e.g., prescribing incorrect diagnosis, giving detrimental recommendations etc. Additionally one has to ensure that AI systems are fair and unbiased. In many contexts, especially in high domains like healthcare and the criminal justice system predictive performance is not enough, hence the need of accountability in the form of explainability, causal analysis, fairness etc.
Reconstruction and Analysis of Historical Social Networks
Historical documents offer a wealth of information about the past including social linkages between people. The aim of this project is to use ancient and medieval texts in various languages to reconstruct social networks. Using medieval and early modern texts in Arabic and English we have been able to reconstruct a social network of more than 20,000 people spanning the globe and all the major civilizations of the world.
Personality Emulation of the Deceased
The death of my father had a profound effect on me. It made me realize that my children will never have the opportunity to interact with him. This has prompted me to start work on this project where the aim is to create simulations of deceased people so that the living can interact with them. In this project not only are the requirements for creating such a system explored but their implications at the intersection of artificial intelligence, natural language processing, psychology and sociology.
Data Mining Pilgrimage Data
Every year millions of Muslims go to the Holy cities of Mecca and Medina to perform the pilgrimage of Hajj. Over the course of just two weeks more than 3 million people are housed within an area less the size of Manhattan. Routing and providing services such a large mass of people is a logistical challenge that can be addressed via the application of machine learning to Big Data. This project was focused on data mining pilgrimage data in order to provide provide insights that can greatly improve the logistics of Hajj.
Modeling Human Behavior in Massive Online Games
This work was done as part of the Virtual Worlds Observatory project which was a multi-disciplinary collaboration between University of Minnesota, University of Southern California, University of Illinois – Urbana Champaign and Northwestern University. This project was focused on the use of machine learning to model human behavior in massive online games to study phenomenon like team formation, trust formation and evolution, predictive models for demographic characteristics (age, gender etc.) and behavioral characteristics (leadership, political affiliation, deviancy etc).
Computational Models of Trust in Massive Online Games
Computational trust refers to the mediation of trust via a computational infrastructure. In this project questions related to trust in complex social environments represented by Massively Multiplayer Online Games (MMOGs) are explored. The main emphasis is that trust is a multi-level phenomenon both in terms of how it operates at multiple levels of network granularities and how trust relates to other social phenomenon like homophily, expertise, mentoring, clandestine behaviors etc. Social contexts and social environments affect not just the qualitative aspects of trust but this phenomenon is also manifested with respect to the network and structural signatures of trust network.
Clandestine and Deviant Behaviors in Massive Online Games
Gold Farming refers to a set of inter-related activities in online virtual spaces especially in Massively Multiplayer Games where certain players engage in repetitive activities to gain virtual commodities which they sell to other players. Game administrators actively ban Gold Farmers who have to hide their activities from the game admins, who are akin to law enforcers. In our research we compared Gold Farmers with their offline criminal counterparts. It was discovered that the behaviors of Gold Farmers is very similar to real world criminals in that their social networks tend to be quite similar.
Machine Learning for Historical Census Data
Before the US instituted Social Security numbers as a way to uniquely identify all US citizens, there was no way to identify people across US censuses. The aim of this project was to develop a machine learning model and framework to link people across US historical census data from 1850 to 1910 e.g., linking and then tracking the same person from 1850, 1860, 1870, 1880, 1900 to 1910 censuses.
Data Mining ICD Data
The aim of this project is to extract useful insights from ICD (Implantable Cardioverter-Defibrillator). This is the project that I did at Boston Scientific in St. Paul Minnesota. Time series data for tachycardia/tachyarrhythmia and other types of arrhythmia was extracted and different pattern set mining techniques were applied to it. This project hit close to home later on because my father was implanted with the device that I worked on.