About

I'm a third-year B.Tech student pursuing Computer Science with specialization in Bioinformatics at VIT Vellore. My research focuses on machine learning, with an emphasis on real-world applications. In particular, I'm interested in AI safety, attention techniques to accelerate learning and inference, approaches toward interpretable deep learning, and methods for adversarial machine learning.

I'm a machine learning researcher at DSC-VIT (formerly known as GDG-VIT), working as machine learning lead on a motion simulation and wearable tech startup called Motus Simulation, under the roof of Creation Labs. Also am the AI6 City Ambassdor for Vellore.

Work Experience

Deep Learning Intern
Symphony AI
May 2018 - July 2018
Mentored by: Vivek Vaidya

I worked on Interpretability of deep learning models and worked on implementing research papers which provided SOTA on publicly available highly imbalanced datasets for sentiment classification, which was later used in proprietary healthcare data for staging purposes. Then worked on ETA of death of a patient by using survival regression techniques. At last, worked on autoencoders for clustering purposes to tackle unsupervised learning situtations.

Models built for Classification:

  1. Convolutional Neural Networks for Sentence Classification by Yoon Kim, EMNLP 2014
  2. An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling by Shaojie Bai et al.
  3. Attention Is All You Need by A Vaswani et al, NIPS 2017
  4. Relational recurrent neural networks by DeepMind's by Adam Santoro et al.
  5. Hierarchical Attention Networks for Document Classification by Microsoft Research's Yang et al.

Model built for Regression:
  1. Weibull Time To Event Recurrent Neural Network by Egil Martinsson.

Model built for Clustering:
  1. Deep Clustering with Convolutional Autoencoders by Guo et al, ICONIP 2017

Papers on Interpretability:
  1. Learning how to explain neural networks: PatternNet and PatternAttribution by Google Brain Kinderman's et al also from TU Berlin, ICLR 2018
  2. The attention papers mentioned above.

Tools of choice: Python, Keras, Pytorch, TensorFlow, Pandas, scikit-learn.


Machine Learning Research Intern
CSIR-CDRI
Jan 2018 - Mar 2018
Mentored by: Tomin James

Had to implement few research papers related to the prediction of Onset and Event Epileptic seizures, using a really huge dataset.

  1. Automated diagnosis of epileptic EEG using entropies by Acharya et al, Biomedical Signal Processing and Control.
  2. Comparative study on classifying human activities with miniature inertial and magnetic sensors, Altun et al, Pattern Recognition.

Tools of choice: Python, Keras, Pytorch, Pandas, scikit-learn.

Projects

  • Sometimes Deep Sometimes Learning: Machine Learning Experiments with scikit-learn, Deep learning with Keras, TensorFlow and Pytorch. Data Science examples for various datasets and competitions from Kaggle and Analytics Vidhya.

  • Anti Spam Filter: Anti Spam Filter, a spam filter which uses a model made out of MultinomialNB algorithm from scikit-learn snake to classify hostel complaints against spam, tested with various languages and abusive words.

  • kaala: A recommendation software, for suggesting crops to Tamil Nadu farmers.

  • rpi-car: A Raspberry Pi controlled car with TensorFlow Object Detection and GUI built on PyQt5.

  • PDFtools: Scripts to merge and split PDFs, merge images into PDFs.

  • YDL: YouTube Downloader for downloading audio files, video files, playlists of both video and audio at the preferred quality for free.

Achievements

  • fast.ai International Fellow, Pytorch version - March 2018 | USF Data Institute
  • kaala Won the second place at VIT Vellore's Software Engineering Hackathon.
  • Ranked 5186th (as of July 2018) out of 150,000+ Data Scientists at Analytics Vidhya.

Volunteer Experiences