Oracle Labs

The goal of the internship was to scale topic models which are provides highly versatile tools for modelling and representing complex problems like for modelling user interest and behaviour, to describe documents and networks. However, modern data analysis requires computation with massive datasets and scaling inference for topic models to these sizes is challenging. Moreover many real-world problems fit more naturally in an online learning framework where small bunches of documents arrive in a stream and has to be discarded after one look. In this talk, we explore various streaming algorithms for topic models present in the literature that can analyzes massive collections of document. Also we present the extension of the the newly proposed SCA learning for topic models to this streaming case. This method is embarrassingly parallel and can leverage inherent sparsity present and faster sampling techniques like alias method, while possessing similar predictive power.

Intel Strategic Labs

The goal of the internship was to improve training phase of certain product, which comprised of multiple steps. In this regard I developed a Markov Decision Process based methodology to optimize training phase by selecting optimal steps and the MDP model was learnt based on old logs.

University of Auckland

I got an opportunity to work with full scholarship at the power electronics group at University of Auckland, New Zealand, which is famous in the field. In particular my task was to design a minimal sized system for low power applications. I came to know how doing research in one topic is actually so multifarious that it requires knowledge of many disciplines. For implementation of controller Cypress Programmable System on Chip (PSoC) as per my advisors’ recommendation was used (I found it quite a nice tool for development and became interested in how they were designed, in fact in the entire field of reconfigurable systems). Assembling the circuit based on a novel tuning idea, not only helped me further understand how the system responded to changes in different parameters in a practical environmental setting, but also provided the joyful exaltation of witnessing my troubleshot design work as desired. This experience culminated with the publishing of a paper.

Carnegie Mellon University

10-701 Machine Learning

I got the opportunity to TA this graduate class. This course gives students a thorough grounding in the methods, mathematics and algorithms needed to do research and applications in machine learning covers topics ranging from concept learning, decision trees, neural networks, linear learning, active learning, estimation & the bias-variance tradeoff, Bayesian learning, the MDL principle, Naive Bayes classifier, Bayes Nets & Graphical Models, the EM algorithm, Hidden Markov Models, MCMC sampling, K-Nearest-Neighbors and nonparametric learning. My responsibility comprised of design and solutions of exam and homework problems, grading, and holding recitations and office hours.

18-202 Mathematical Foundations of Electrical Engineering

I got the opportunity to TA this undergrad class. This course covers topics from engineering mathematics that serve as foundations for descriptions of electrical engineering devices and systems. My responsibility comprised of design and solutions of homework problems, grading and holding office hours.