Page 1 of 1
(585) 309-0119
srinathobla@mail.rit.edu SRINATH OBLA github.com/srinathos
linkedin.com/in/srinathos
EDUCATION
Rochester Institute of Technology MS in Computer Science October 2019
Focus: Machine learning, Pattern Recognition, Computer Vision, Computational Geometry, Parallel Computing GPA 3.85
University of Mumbai, India BE in Computer Engineering June 2016
Focus: Image processing, Big data analytics, Data warehousing, Software development life cycles and tools
LANGUAGES AND TECHNOLOGIES
• Languages: Java, Python (Proficient), MATLAB, C++ (Intermediate), JavaScript, SQL, HTML, XML, PHP, Bash, R (Familiar)
• ML/CV: PyTorch, Keras, TensorFlow, Caffe2, sk-learn, OpenCV, dlib, Pandas, Numpy, Point Cloud Library, MeshLab
• Frameworks and Tools: Parallel Java 2, Hadoop, Android, Git, Raspberry Pi, AWS, Django, Docker, CUDA, CSS
RESEARCH AND EXPERIENCE
Thesis – Python, PyTorch, Matlab Rochester Institute of Technology Ongoing
• Enabled deep neural networks to process encrypted data without decryption using Homomorphic encryption (HE).
• Discovered a novel approach using weighted polynomial approximations to construct activations functions for HE.
• Developed scripts to analyze hidden layer outputs in CNNs, train multiple networks and record network statistics.
• Proficient in transfer learning, data augmentation, hyperparameter tuning and hypotheses testing.
• Awarded a $2000 scholarship by TSO Logic, an AWS Company for contributing to the field.
Research Assistant – PyTorch, Matlab Rochester Institute of Technology Dec 2018
• Identified that modern activation functions have bounded derivatives that contribute to good performance.
• Developed multiple custom activation functions and tested against datasets like FMNIST and CIFAR-10.
• Achieved an accuracy of 92.33% on CIFAR-10 using custom activation functions with the identified property.
Teaching Assistant – OpenCV, C++, MATLAB Rochester Institute of Technology May 2018
• Provided teaching assistance on topics including Image/video processing, motion tracking, depth estimation.
• Graded exams and developed future homework and projects to be assigned in the course.
Intern – Java, Android Yu Televentures March 2015
• Performed reliability and performance tests on system, third-party applications and the Android OS pre-launch.
• Provided the developer community with documentation and developer access support.
PROJECTS
Differentiating between music and advertisements – Python, Pandas, sklearn May 2019
• Developed an MLP classifier to identify audio playing in radio streams as either music or advertisements.
• Achieved an accuracy of 93.25% and an F-score of 0.93 using features engineered from raw audio streams.
• Created a labelled dataset of 213,000 samples from music and advertisement radio audio.
Automatic blurring of faces in videos – dlib, OpenCV March 2019
• Discovered faces in videos using a Convolutional Neural Network and preserved privacy by blurring them.
Segmenting, Classifying and Parsing handwritten math equations – Python, sklearn May 2018
• Developed a model with an accuracy of 84.52% that segments a group of handwritten strokes into meaningful symbols
• Achieved a classification accuracy of 87.9% by training a random forest classifier on 101 symbol classes.
• Engineered 36 features (bounding-box, histogram) determining spatial relationships between symbols.
Monitoring speed and direction in vehicular traffic flow – OpenCV, C++, MATLAB Dec 2017
• Built a computer vision model to identify and track vehicles in a parking lot with pedestrian traffic.
• Estimated speed of vehicles over multiple frames and reported detected anomalies.
3D Object Reconstruction using depth sensors – C++, PCL, OpenNI, Meshlab Dec 2017
• Captured dense point clouds of the object from multiple angles using a depth sensor and a custom turntable.
• Created a 3D model by aligning clouds using iterative closest points (ICP) and stitching them.
Fingerprint detail extraction using Mobile phone cameras – MATLAB May 2017
• Processed fingerprints from RGB images using morphology and color space transformations.
• Extracted ridge and valley fingerprint features (minutiae) by processing skeletonized fingerprint image.