Prahalathan Sundaramoorthy

Prahalathan Sundaramoorthy

Master's student at the University of Southern California. Specializing in machine learning and big data

Work Experience

HP Labs

Research Intern

May 2020-Present, Palo Alto, United States

  • Devised an end-to-end Heart Rate calculation model from Photoplethysmography (PPG) time-series sensor data in Python. Reduced mean error by 30% through spectral filtering and peak validation; collaborated with production team for porting model to C for release.
  • Achieved 80% accuracy in predicting cognitive load of individuals through PPG wave morphology. Trained 1D Convolutional Neural Network to capture shape of data from 450 participants on AWS EC2.
  • Optimizing aforementioned models by integrating data from Inertial Measurement Units.

Independent Research

August 2018-December 2018

  • Developed ActiveHARNet, a Bayesian Convolutional Neural Network capable of active learning to update model based on real-time unlabelled data for Human Activity Recognition.
  • Trained model online and improved mean accuracy by ~25% using 50% incoming data.

Teknuance Info Solutions Pvt. Ltd

NLP & Deep Learning intern

June 2018-August 2018, Chennai, India

  • Examined deep learning techniques such as seq2seq LSTMs, Word2Vec, GloVe, etc. for performing NLP tasks – text summarization, topic modeling on business data.

Solarillion Foundation

Research Assistant

April 2017-June 2018, Chennai, India

  • Led a team of four to develop a Human Activity Recognition system from smartphone sensor data with 43 Million samples. Tested various CNN and RNN ensembles and designed HARNet.
  • Equipped HARNet with support for on-device incremental learning resulting in an increase in accuracy of ~35% of least performing end-user.
  • Devised a low-cost system for Non-Intrusive Load Monitoring on Raspberry Pi employing Ensemble Machine Learning (Extremely Randomized Trees) with inference time of 400ms and 86% accuracy.