1. Research
Sep’18-Now Medical Imaging & Deep Learning Gevaert Lab, Stanford
- Worked on imaging bio-marker segmentation and neural autoen-
coder feature analysis using Keras models.
- Developed a model incorporating registered patch based approach
for hassle-free training in disease classification.(∼ 3% error)
- Designing a Multi-class model for neural images with molecular
characteristics for early diagnosing dementia and Alzheimer’s.
Apr-Now Meta-gradient Learning| Reinforcement Learning Stanford
-Proposed and evaluated primal-dual meta-objective alleviating the
double sampling. Better learning than baseline in toy MRP and ATARI
Work Experience and Internships
Jun-Sep’18 Summer Intern NVIDIA Corporation| Santa Clara, CA
-Developed a Deep learning solution to optimize GPU testbench &
coverage( 20X impr.). Neural net model in PyTorch for prediction
-Trained feature segmentation algo improving control (Turing GPU)
2016-May’17 Research Assistant Indian Institute of Science|CGRA Team
-Designed a reconfigurable Vector Processor for streaming kernels.
( 3X perf over scalar processors)
-Synthesized real time Face recognition Neural Net in C++, Python
Projects
Sep-Dec’18 Artificial Intelligence| Robotic Digit Mimicking Stanford
-Developed a system to write a digit from learning actions at all states
-Devised algorithm (in Python ) to trace out simple digits (Acc. 98%)
Sep-Dec’18 Generative Models| Auto Doodle generation Stanford
-Generated intricate doodles based on simple pen strokes and mini-
mal sketching (utilizing Transfer Learning ).
-Trained CycleGAN transforming hand sketch to/from doodle art
Apr-Jun’18 Deep Learning| Neural Net Approaches to DNA Denoising Stanford
-CNN approach (U-Net architecture) predicting entire denoised DNA
sequence (0.05% error on reference sequences)
-RNN model predicting localized nucleotide substitution/deletion;
model in Keras, TF (2.8% error/seq.)
Apr-Jun’18 Conv Net| Artwork Classification and Style Transfer Stanford
-Designed high accuracy artistic media & emotion classification so-
lution ; model in TensorFlow & Transfer Learning(close to VggNet)
-Performed style transfer transforming media/emotion of images.
Sep-Dec’17 Machine Learning| Supervised autonomous driving Stanford
-Formulated end-to-end steer and throttle driving control from
recorded raw images, trained CNN ( TensorFlow ) with LSTM ends.
-Worked on improving "performance metrics" to increase max speed
without offshoot & image processing NVIDIA architecture.
Achievements
2016 CDNLive Best Paper Award
Functional safety analysis verification solution
2015 Best Project APOGEE
Intelligent lighting control demo mechanism
2015 Runner up Paper APOGEE
Robust Iris segmentation hardware module
Electives/Online Courses
Creative Thinking|AI in Imaging|Computer Vision| Data Science
Abhishek
Roushan
Stanford, CA
aroushan@stanford.edu
Interested in AI/ Deep learning
applications in data & imaging
Education
Stanford University
MS in Electrical Engineering |
Depth: Software & Hardware
Systems| Jun 2019 | GPA:3.74/4.00
BITS Pilani, India
B.E in Electronics & Instrumentation
Depth: Computer & Hardware
systems| Jun 2016| GPA: 9.45/10.00
Skills
Languages: Python, C/C++, OpenCV,
MATLAB, Java, Scala, Perl, LATEX
DL Lib.: PyTorch, Tensorflow, Keras,
Google Cloud Platform
WebDev : HTML, JS
Simulation : ModelSim, ASM
Other: MySQL, Git, MapReduce
Miscellaneous
Coursework
Machine Learning/Deep Learning
Artificial Intelligence/ Reinf. Learning
CNNs/Generative Networks
Virtual Reality
Computer Systems
Computer Vision
Links
Github: https://bit.ly/2C5Bzff
LinkedIn: https://bit.ly/2QCuua0
Quora: https://bit.ly/2IQuwZr
Publication
Compact modeling of a parabolic
cross section nano-FinFET, IEEE
(https://bit.ly/2OL9a54)
Strengths
Dilligent • Quick Learner • Proactive
Life Ideology
"If you don’t change the way to look
at things, things won’t change!"