Syed Tousif Ahmed
Syed Tousif Ahmed

Hi, I'm Syed!

I was born in Dhaka, Bangladesh. A very sweet place, has very beautiful rains and 2-3 hour long traffic jams. I was there for the majority of my childhood and teenage years. In 2011, I moved to Kolkata, India to complete my Grade 11 and 12. Life there was quite dynamic and the place turned out to be very enriching for my personal development. Kolkata gave me life long friends and bitter and sweet memories. After Kolkata, I moved to Rochester, NY, USA and completed my Bachelors in Computer Engineering from Rochester Institute of Technology. I spent a little more than a year at NVIDIA as a Software Engineer and contributed to PyTorch. I am currently pursuing a Ph.D in Electrical and Systems Engineering at UPenn, where I'm advised by Dr. Andre DeHon.

Academically, I love to investigate high performance computing, machine intelligence, digital logic design, compilers and cryptography. Sounds too nerdy? Here are some more things – I love singing, playing the guitar, piano, percussion, philosophy, poetry (middle eastern Sufi poets especially!), travelling, reflecting, sleeping, criticizing, eating, observing people’s face and a lot of other things that you can discover about me if we ever spend some time together.

Click the links above (or on the hamburger menu if you are using a mobile device) to learn more about my career.


  • 08/02/2019 Left job at NVIDIA to pursue Ph.D at UPenn

  • 06/10/2018 Started full-time job at NVIDIA

  • 12/01/2017 Visiting UPenn

  • 11/17/2017 Gave TensorFlow tutorial at Western New Image and Signal Processing Workshop 2017

  • 11/17/2017 Presented poster on systolic arrays at Western New Image and Signal Processing Workshop 2017

  • 10/31/2017 Presented poster on deaf/hard of hearing tech at ACM Assets 2017

  • 08/28/2017 Started senior year Fall 2017

  • 05/31/2017 Started Internship at NVIDIA in Santa Clara, CA as a Deep Learning Software Intern

  • 05/09/2017 Oral Session on Real Time American Sign Language Video Captioning using Deep Neural Networks at GTC 2017

  • 01/23/2017 Started Spring 2017


University of Pennsylvania | Philadelphia, PA, USA
Ph.D in Electrical and Systems Engineering
Expected Graduation: May 2023

Rochester Institute of Technology | Rochester, New York, USA
BS in Computer Engineering
Minor in Mathematics
Graduated: May 2018

Oaktree International School | Calcutta, India
International Baccalaureate Diploma
Math HL, Physics HL, Chemistry HL, English SL, French SL, Economics SL
Extended Essay: Mathematics (Lucas-Lehmer Algorithm vs Trial Division)
Graduated: May 2013

South Breeze School | Dhaka, Bangladesh
Edexcel IGCSE
Graduated: May 2010

Work Experience

Deep Learning Software Engineer | June 2018 - August 2019
NVIDIA Corporation | DL Frameworks Team | Santa Clara, CA
- Worked on different projects for PyTorch in the Deep Learning Frameworks Team
- Resolved several bugs and performance issues on both CPU and CUDA. Improved performance of several Random Number Generation kernels by upto 3x
- Participated in several code reviews and collaborated with third party contributors
- A full list of contributions can be seen here
- Managed Continuous Integration test environment and released NVIDIA GPU Cloud PyTorch docker container on a monthly basis
- Got into top 100 PyTorch contributors list within 7 months of joining the team and currently a module level maintainer of the CUDA backend
- Languages/Technologies used: C++, Python, CUDA, PyTorch, Docker

Research Assistant | September 2016 - May 2017; August 2017 - Present
Center on Access Technology, NTID, RIT | Rochester, NY
- Worked on developing a research tool that investigates the different needs of the deaf/hard of hearing population
- Integrated Automatic Speech Recognition Engines into the tool
- Worked on Video to Text model for American Sign Language
- Languages/Platforms used: TensorFlow, Android Development, Node.js

Deep Learning Software Intern | May 2017 - August 2017
NVIDIA Corporation | DL Frameworks Team | Santa Clara, CA
- Worked on different projects for Caffe2 and Tensorflow in the Deep Learning Frameworks Team
- Designed and implemented Universal Framework Format Format (UFF) Converters for TensorFlow and Caffe2, released in TensorRT 3.0 RC
- Wrote Sequence to Sequence Framework for Caffe2
- Analyzed performance of Caffe2 kernels for Seq2Seq models and made optimizations
- Languages/Technologies used: C++, Python, Protobuf, CUDA, Caffe2, TensorFlow

Research Assistant | September 2015 - December 2015; August 2016 - May 2017
FETLab, GCCIS, RIT | Rochester, NY
- Assisted in research projects in the area of human-computer interaction. The focus of the research projects was on wearable and mobile computing and how to make personal fabrication technology such as 3D printers, laser cutters, and CNC routers easier to use by non-experts
- Wrote code for a kinect and projector based project for augmenting fabrication in laser cutters, 3D printers etc.
- Built an automatic speech recognition system that classifies sounds of actions on everyday objects
- Wrote Android Wear applications
- Languages/Technologies used: Python, C#, Java, Android Development

Deep Learning Engineering Intern | June 2016 - August 2016
NextDroid, LLC (Startup) | Cambridge, MA
- Wrote neural network models for road image segmentation for a semi-autonomous/self-driving car at NextDroid LLC., a startup based in Cambridge, MA
- Wrote image segmentation web interface for mass data collection, that decreased data collection cost by 60%
- Wrote models in caffe, torch and tensorflow
- Wrote unit tests in tensorflow
- Wrote custom operation layer in tensorflow
- Platforms/Frameworks used: Caffe, Tensorflow, Torch, CUDA, NVIDIA Jetson TX1, NVIDIA DRIVE PX
- Languages used: Python, C++, Lua

Computer Vision Developer (Co-op) | January 2016 - May 2016
Ahold Delhaize | Quincy, MA
- Worked on computer vision related projects at the Propulsion Labs at Ahold USA
- Used tensorflow and caffe to do transfer learning for proof of concept on product image recognition
- Used OpenCV for data augmentation and production of synthetic images
- Used OpenCV and OpenGL ES to make an augmented really iOS app that gives a location-aware shopping experience. Wrote UI elements and their functionality in Redux.js and React.js
- Languages used: Python, C++, Objective-C, JavaScript

Computer Vision Research Assistant | June 2015 - December 2015
Discover Lab, School of Media Sciences, RIT | Rochester, NY
- Developed, debugged, and optimized an augmented reality app, called RocreadAR for a research project aiming at integrating different media for publishing and communication
- Wrote Image Processing algorithms in OpenCV and Python to enhance detection and tracking of target images
- Received NSF I-Corps Funding and assigned as the Student Team Leader to commercialize prototypes
- Technologies used: OpenGL, OpenCV, Unity3D, Vuforia SDK, Wikitude SDK, Git, Android, iOS, Google Glass


User Experiences When Testing a Messaging App for Communication Between Individuals who are Hearing and Deaf or Hard of Hearing
Lisa Elliot, Michael Stinson, Syed Ahmed, Donna Easton
Assets 2017

Using Automatic Speech Recognition to Facilitate Communication Between an Individual who is Hearing and One who is Deaf or Hard of Hearing
Michael Stinson, Syed Ahmed, Lisa Elliot, Donna Easton
Assets 2017

Further Investigations into Round Touchscreen Wristwatch Interaction
Dhwanit Mehta, Patrick C Shih, Syed Ahmed, Daniel Ashbrook
Abstract: While circular touchscreen wristwatches have become a reality, their interfaces are still based on a square-screen paradigm. In this paper, we explore an interface for round watches based on placing selection items around the edge of the screen rather than in a list. We present the results of a study comparing our round interaction with the standard list interface, and of a study investigating the efficacy of our interaction in an eyes- free scenario.

FETlab, GCCIS, RIT | Spring 2017
One Shot Learning for Acoustic Recognition
Syed Ahmed
Western New York Image and Signal Processing Workshop 2016

Syed Ahmed, Suresh Jothilingam, Yogesh Jagadeesan, Elena Fedorovskaya
Discover Lab, School of Media Sciences, RIT | Fall 2015

Augmented Fabrication
Syed Ahmed, Sourabh Kulhare, Ameya Lonkar, Daniel Ashbrook
Abstract: In this work, we used a kinect and a projector to continously project the image of an object onto itself. This application can then be used on a laser cutter where the projector would be mounted on the ceiling above the laser cutter. A designer can then visualize the design in real time on top of the object and make changes before sending the final design to the laser cutter.

FETlab, GCCIS, RIT | Fall 2015
Out of trial division algorithm for finding primes and Lucas-Lehmer algorithm for finding Mersenne primes, which algorithm would yield a big prime number faster?
Syed Ahmed
International Baccalaureate Extended Essay Spring 2013

Patterns in Complex Numbers
Syed Ahmed
International Baccalaureate Fall 2012

Probabilities in Games of Tennis
Syed Ahmed
International Baccalaureate Spring 2012


TensorFlow Tutorial
Syed Ahmed
Invited Speaker, Western New York Image and Signal Processing Workshop 2017

Real Time American Sign Language Video Captioning using Deep Neural Networks
Syed Ahmed
NVIDIA GPU Tech Conference 2017

Transfer Learning on Tensorflow in 30 minutes
Syed Ahmed
TensorFlow Boston Meetup, May 7 2016


Teaching Assistant, CMPE 679 Deep Learning | Dr Ray Ptucha, RIT, Spring 2018
Author homework assignments
Hold recitations

CMPE 789 Special Topics (Deep Learning) | Dr Ray Ptucha, RIT, Spring 2017
Authored homework assignment on TensorFlow.

Supplemental Instruction Leader | RIT, Spring 2015
- Supplemental Instruction Leader for Academic Support Center at RIT
- Conducted an hour long study session twice a week, through the last day of classes per semester, to guide students with historically difficult courses (courses with high rates of D, F and withdrawal)
- Planned and marketed SI Sessions through weekly session announcements in class and through email
- Engaged with faculty partner and devised SI session strategies using different learning techniques
- Attended weekly staff training, evaluated and reflected on SI sessions conducted by colleagues


Real Time American Sign Language Video Captioning
- Implemented Sequence to Sequence Neural Network for translating American Sign Language video to text.
- Presented at GPU Tech Conference 2017. See Talk above.

Savage Valhalla: Building a Fast Autonomous Car
Syed Ahmed, Jeff Barker, Christopher Ranc
- Autonomous RC car that can follow a white track bounded by two black lines on the side
- Awarded second place on Imagine RIT Autonomous Car Competition Spring 2017
- Used an ARM K64 board and concepts from microcontroller programming such as UART, PWM, etc. See details in the paper.

Open source contributions in Deep Learning Research
- Caffe2 - Solved several bugs in Seq2Seq. Commit #35dc34
- TensorFlow - Solved bug in Android Demo. Issue #1371
- TensorFlow - Implement Max Unpooling Op - Issue #2169
- elab/Torch7-profiling -Solved bug-Commit #7fdb7af and #0e64c08
- elab/ENet-Training - Improved code - Pull request #9

An Augmented Reality android app that gives an interactive resume reviewing experience. Made using Unity3D, Vuforia SDK, and Android Studio. Works on this resume.

Click Wars - RIT iOS App Challenge Hackathon 2015
Syed Ahmed, Towhidul Chowdhury
A game based app called “Click-Wars” that uses face detection and bluetooth to connect multiple players to play a game of who can click each others face faster.