Skip navigation

Research

Most of my past research has been related to some facet of Machine learning

Computationally understand meaning of a concept

Advisors: Prof. Luis Francisco-Revilla and Prof. Brenda Berkelaar, UT Austin
Graduate research assistant

Built a prototype for a system that attempts to computationally understand the meaning of a social contract and how it changed with time. We analyzed historical newspapers by building distributional semantic models.

Scaling neural language models

Advisor: Prof. Yoshua Bengio, Universite de Montreal
Internship

The summer after my junior year, I interned at LISA, the machine learning lab of Universite de Montreal. I worked on different aspects of scaling up neural networks.
  • I worked on an idea called conditional computation, where additonal neural networks called gaters determine which units in the original network should be computed corresponding to a particular input. We can exploit the sparsity of activated units, by calculating only for these units and save computation. I built models based on this with different architectures and investigated their performance.
  • When dealing with very large dictionaries for neural language models, the normalization factor in the output softmax layer becomes intractable. I implemented Hierarchial softmax and Noise contrastive estimation to overcome this issue and compared their efficiency against training with a regular softmax layer.
  • I also built a system to generate n-grams on the fly (at run time) for very large datasets. For such datasets, it becomes infeasible to generate and store all possible n-grams because of the memory size required for that.
  • The development was done in Python, using the libraries Theano and Pylearn2.

Learning from text commentary of the game of cricket

Advisor: Dr. Vijaya V. Saradhi, Assistant Professor, IIT Guwahati
Undergraduate thesis

  • Exploring strengths, weaknesses and playing strategies of cricket players and their relationships with external factors like weather, batting order, etc.
  • Investigating the usefulness of Canonical correspondence analysis to ordinate data points along gradients of important external variables, to come up with a low dimensional triplot describing the traits of players and the influence of external factors
  • Extracted features from text commentary of the game using a combination of text mining techniques

Statistical methods for better network traffic analysis

Advisor: Dr. Vijaya V. Saradhi, Assistant Professor, IIT Guwahati
Summer research project

  • Devised flow based representations of network traffic, where each flow is made a node and similarity between flows constitutes an edge, as an alternative to overcome the deficiencies of traditional packet-based traffic dispersion graphs (TDGs)
  • Analyzed how well these representations captured traffic flows, and their relationships with TDGs using Canonical correlation analysis