Naveen Mathew Nathan - Personal website

Summary

I’m a (data) scientist with 6.5 years of industry experience in machine learning R&D. I currently work on fundamental scientific problems, delivering insights from experimental data. I’m deeply interested in mathematical foundations of machine learning and deep learning. I’m also interested in creating data pipelines and automatic machine learning (including unsupervised learning, which is often ignored in popular AutoML frameworks). In the past I worked with client delivery teams to create automated pipelines for supervised and unsupervised learning (details).

Product/Solution Development

Short presentations on my R&D work in industry can be found below:

• Automated pipeline for binary classification [View presentation] [View repo on GitHub]
• Automated pipeline for clustering and basket anlaysis [View presentation] [View repo on GitHub] [View use case presentation]
• Pipeline for natural language understanding and predictive modeling [View presentation] [View repo on GitHub] [View use-case presentation][Example notebook]

Research

My research interests:

• Unsupervised learning and crowd-sourcing for exoplanet candidate identification [Motivation] [Blog post]
  • Why do we need unsupervised learning and crowd-sourcing for exoplanet candidate identification? [Blog post]
  • An old version of the application is live! [Link to the application]
    • Demo details: username: user1, password: pass1
• Reinforcement learning for traffic control [Blog post]
  • Applied for a US patent, pending approval
• Information theory based redefinion of binary cross-entropy to define a new link function in GLM (binary classification) [Mathematical Details]
• Applications of machine learning and deep learning in finance
  • Applied classical machine learning models and AutoML to predict the direction of movement of bitcoin price in real-time [PDF Report] [Technical Presentation]
  • Used deep learning to predict the direction of movement (in next 60 seconds) of VWAP of most significant levels in limit order book [PDF Report] [Technical Presentation]
  • Scaled up terabyte scale financial data processing using multiprocessing [Article on scaling consecutive row diff using Python]
  • Deep transfer learning for limit order book analysis [PDF Report] [Technical Presentation]

Course Work

To know more about my course work and to get a copy of my personal course notes (LaTeX), please visit this page. Important things to note:

• The R Markdown scripts can be compiled to generate PDF outputs. I take extra care not to upload PDF files to my course repository. Please DO NOT contact me for assistance on compilation to PDF
• Please DO NOT contact me for feedback on courses or professors. Most importantly, please DO NOT contact me for assignments. I will not entertain any violation of honor code of UIUC before
• Please DO NOT contact me for the actual slides used in courses. If you are interested in the program please visit the Statistics department website

Knowledge Sharing

• I share my knowledge through my blog on Medium.com. My main focus is on the mathematical and statistical fundamentals of machine learning and deep learning. I also share my practical experience by explaining how mathematics helps in solving real-life problems. I update links to my blog posts on this page
• I recently started putting together all my blogs in the form of a loosely ordered book. Most of the pages will be empty except the chapter on Generalized Linear Models / Linear Regression