I am a Ph.D. student in Center for Data Science at New York University. Previously, I was a Research Engineer at Adobe Research in BigData Experience Lab, India. I completed my Integrated Masters from Indian Institute of Technology Delhi in Mathematics and Computing (2015). I was fortunate to be mentored by Prof. Amitabha Bagchi and Prof. Parag Singla for my thesis project. I interned at Adobe Research in the summers of 2013 and 2014. Before that I spent a summer as a visiting researcher at BME, Hungary.
I am broadly interested in machine learning for structured data such as graphs, sequences and time series. In particular, I have worked on learning features for nodes in a social network which used concepts from statistical language modeling. Another work focused on leveraging statistical techniques from Survival analysis for predicting time to opening an email using sequence of user activity on emails. I have also used neural network variants for sequential data such as recurrent neural networks and memory networks for forecasting time series of web metrics and for predicting future performance of students on online assessments. More recently, I have been exploring bandit algorithms for learning to rank.
Selected Research Projects
Representation learning in graphs
In recent years, there has been significant work on learning representations of network data, especially using neural networks. These representations can be used for tasks such as inferring properties of entities within the network. Prominent application domains include social networks, knowledge bases and biological interaction networks. We investigated a method to learn representations of users in social networks from the activity traces of users. Current techniques rely on the underlying network structure for learning such representations, which might not be known reliably. The proposed technique can be used to extract features for prediction tasks in social networks such as link prediction. arXiv link
Survival analysis for Emails
In email marketing, it is typical to observe moderate to very low open rates among customers. In a recent work accepted for publication at WSDM 2018, we consider the problem of predicting whether a customer will open an email and how much time will be taken for opening the email. Survival analysis offers a framework to model time to event data where event in our case corresponds to an email being opened. A mixture model based approach is used that accounts for the low open rate characteristic of email data.
Online learning to rank
Many of the modern web applications present users with a list of items which they can choose from. Examples includes web search results, product recommendations on ecommerce sites, news from friends on social media sites and so on. With more data being collected on users interacting with the list of items, the problem of personalizing these to interests of the users becomes feasible to tackle. We expore the task of interactively learning user interests with an objective of minimizing abandonment of lists i.e. user finding no item of interest.
Please see the Publications page for complete list of projects.
Other Interests I enjoy playing Basketball, quizzing, listening to music and (trying) to play guitar.
Feel free to contact me via email.