Layerwise align the neurons across different neural nets by OT and then average their parameters.

Representing each entity as a distribution over contexts endowed in a ground space.


Disclaimer: The ‘2020 Reflection(s)’ refer to only my own personal views! (also, serves as an amusement)

September 2018 – February 2019
Menlo Park, California

Research Intern

Facebook AI Research

Worked on building non-compositional embeddings for application in text representation and generation.

  • 2020 Reflection: NLP is a super cool area, and I am really fascinated by linguistics & how languages evolve. But, I need a short break from NLP research!
  • Bonus reflection : No rush anyways, there is still some time until the septillion-parameter language model gets brute-forced efficiently implemented ;)
May 2016 – July 2016
Kyoto, Japan

Research Intern

Kyoto University

  • Developed a training mechanism for Generative Adversarial Networks (GANs) using entropy regularized Wasserstein distances, guided by Marco Cuturi.
  • Utilized Large Margin Nearest Neighbors (LMNN) for learning the ground metric. Implemented the system in Chainer, with the architectural inspirations from DCGAN.
  • 2020 Reflection: Missed making it work before Wasserstein GAN :/ Nevertheless, what I learned about optimal transport, eventually sparked the core ideas for my next two papers.
  • Bonus reflection: Indebted to Honda Foundation for sponsoring this visit and to Marco Cuturi for teaching me about optimal transport.
November 2015 – January 2016
Bangalore, India

Research Intern

Xerox Research Centre

  • Developed prototype of a multimodal trip planning system that integrates dynamic ridesharing with scheduled transportation services.
  • Used k-medoids algorithm to find clusters of landmarks in road network graph. Implemented a variant of hill climbing algorithm & silhouette analysis to find the optimal number of clusters.
  • 2020 Reflection: Interesting things can be done even without deep learning :P
May 2015 – July 2016
West Lafayette, Indiana

Summer Intern

Purdue University

  • Designed and implemented a method to estimate the relevance of reviews using their metadata, with a particular focus on reviews with limited votes.
  • Implemented consumer Rating as a Service (RaaS) architecture and provided a RESTful API for interaction, which were written using Node.js and Express with MongoDB for persistence.
  • 2020 Reflection: Here, I learned what research is and carried out my first research project :)

Recent Posts

The realization It all happened one evening, during the height of the Coronavirus pandemic in Europe. By then, I had really started …

Transcript is a document which contains the list of all the grades you have received until now and some information regarding the …

On the first day of my sophomore year, I found a few lines inscribed on my hostel room wall in light green color. They were: I thank …