Aditya Kusupati


Research Fellow @ MSR India
B.Tech (Honors), IIT Bombay
Major CSE | Minor EE
Email t-vekusu@microsoft.com
CV here | Resume here

Bio
Codes
Publications
Acknowledgments
Thesis and Internship Reports
Selected Course Projects
Positions of Responsibility
Contact

Bio

I am a Research Fellow in Machine Learning and Optimization Group at Microsoft Research Lab - India. I am advised by Dr. Manik Varma and Dr. Prateek Jain. I am working as part of the Intelligent Devices Expedition Project aimed at developing novel Machine Learning algorithms to make Edge Devices intelligent. My current focus is on fundamental machine learning and deep learning algorithms for resource-efficient and general settings.

The latest output of this project, FastGRNN: A Fast, Accurate, Stable and Tiny Kilobyte Sized Gated Recurrent Neural Network, is accepted for publication at NeurIPS'18.

Before joining MSR, I was an Undergraduate at Indian Institute of Technology Bombay with major (honors) in Computer Science and Engineering and minor in Electrical Engineering. My undergraduate thesis, along with Anand Dhoot, under the guidance of Prof. Soumen Chakrabarti, was focused on making Entity-Typing efficient and geometrically sensible using supervised and self-supervised learning in order to help quite a few down-stream applications like Knowledge Base Completion, Fine Type Tagging etc.,

I also worked on a few interesting projects across various domains of computer science. Have a look at them in my CV and here

I spent my second year undergrad summers at Inria, working in Titane Team under Dr. Pierre Alliez working on Stochastic Mesh Metric Generation for 3D modeling and my next summer at American Express Big Data Labs working on making Gradient Boosted Machines effective under guidance of Dr. Vishwa Vinay.

I am one of the first and major contributors of EdgeML, An ML library for machine learning on the Edge, which proclaims 2 KB (RAM) ought to be enough for everybody. Ping me if you think otherwise, *wink*

I am interested in Machine Learning and Systems in general with a special interest in creating novel and generalisable Machine/Deep Learning algorithms in both Resource Constrained and Large Scale settings, particularly for Search, Time Series and Vision along with a fundamental understanding about the algorithms we propose.

Codes

Publications

FastGRNN: A Fast, Accurate, Stable and Tiny Kilobyte Sized Gated Recurrent Neural Network [Abstract] [PDF] [Video][Poster][Code]

Acknowledgments

New Embedded Representations and Evaluation Protocols for Inferring Transitive Relations

Thesis and Internship Reports

Efficient Spatial Representation for Entity-Typing
Poisson Denoising with Dictionary Learning
Enhancement of Gradient Boosting Machines for Big Data
Polymorphic Anisotropic Metric for Surfaces

Selected Course Projects - Open Research Problems

Gloss: A Continuous Contextual Shape Descriptor
Classification in Social Networks based on Activity in Politics
Language Assistant using predictive POS tags

Positions of Responsibility

Mentor at MSR India Summer Workshop 2018: Machine Learning on Constrained Devices, Summer 2018.

Undergraduate Teaching Assistantship, IIT Bombay

One of the only 2 students to be awarded TA of the month award twice during the period of 2015-2017. * represents the courses for which they were awarded.
  • Digital Logic Design Lab*, Spring 2017 - Prof. Supratik Chakraborty
  • Software Systems Lab, Autumn 2016 - Prof. Sharat Chandran
  • Digital Logic Design Lab*, Spring 2016 - Prof. Supratik Chakraborty
  • Computer Programming and Utilisation, Autumn 2015 - Prof. Varsha Apte
  • Computer Programming and Utilisation, Spring 2015 - Prof. Kavi Arya

Organizational Responsibilities IIT Bombay

  • Department General Secretary, 2016-17, CSE, IIT Bombay
  • Internship Coordinator 2015-16, Placement Cell, IIT Bombay

Contact

gmail github facebook linkedin

Many thanks to Shumo Chu and Sumith Kulal for the site inspiration.