I'm Tejas Srivastava, a Computer Science Graduate Student at the University of Pennsylvania.


Tejas Srivastava

CS Grad Student
University of Pennsylvania

Hi, I'm Tejas , a current master's student at The University of Pennsylvania, pursuing Computer and Information Science, working on a variety of applications of Computer Science, Deep Learning, and Data Science from amodal segmentation in Computer Vision to interpretability of Deep NLP Models.

With a background in Software Design, Web Development, and Programming, I have developed an interest in Machine Learning, Deep Learning, and Data Mining, and am currently focusing on these during my graduate studies, along with core Computer Science concepts such as Operating Systems and Distributed Systems. I am a strong proponent of Data-Driven Artificial Intelligence and the positive impact it is making in businesses and personal lives alike, and have pursued various internships and projects at the junction of Software Development and Data Science to build solutions and explore the applications of these technologies to solve real-world problems.

Education

University of Pennsylvaniasubject

School of Engineering And Applied Science

Master of Science in Engineering, Computer and Information Science (MSE-CIS)

Master of Science in Engineering, Computer and Information Science (2019-2021)close
  • GPA: 3.78/4.00
  • Expected Graduation: May 2021
  • Coursework
  • CIS 520: Machine Learning
  • CIS 581: Computer Vision and Computational Photography
  • CIS 557: Programming for the Web
  • CIS 522: Deep Learning for Data Science
  • ESE 545: Data Mining for Massive Datasets
  • CIS 548: Design and Implementation of Operating Systems
  • Graduate Teaching Assistant:
  • CIS 581: Computer Vision under Prof. Jianbo Shi
  • CIS 520: Machine Learning under Prof. Lyle Ungar
  • Graduate Research Assistant, Advisor: Prof. Lyle Ungar
  • World Well Being Project, Explainable Deep Learning Models for Emotion Classification
University of Punesubject

Pune Institute of Computer Technology

Bachelor of Engineering, Computer Science Engineering

Bachelor of Engineering, Computer Science Engineering (2015-2019)close
  • Graduated First Class with Distinction
  • GPA: 3.87/4.00
  • Relevant Coursework
  • Data Structures and Algorithms
  • Object Oriented Programming
  • Linear Algebra and Statistics
  • Discrete Mathematics
  • Database Management Systems
  • Computer Networks
  • Systems Programming and Operating Systems
  • Software Modeling, Design and Development Lab
  • Information and Cyber Security
  • High Performance Computing
  • Data Analytics
  • Data Mining and Warehousing
  • Artificial Intelligence and Robotics
  • Machine Learning
  • Soft Computing and Optimization Algorithms
  • Teaching Assistant:
  • Software Development Lab (2019)
  • Bachelor's Thesis Project: , Comprehensive Developer's Assistant under the guidance of Prof. Mayur Chavan
  • Bachelor's Seminar: Identification of Plant Species from images of plant organs, under the guidance of Prof. Mayur Chavan
  • Activties and Involvements
  • Public Relations Officer and Marketing Head, PICT IEEE Student Branch
  • Member of Coding and Web Development Teams, PICT IEEE Student Branch
  • Member of Technical Team, Pradnya, Impetus N Concepts
  • Member of Design Team, Elevate (InterCollegiate Sports Event)

Projects

Deep Learning
Computer Vision
Research
Semantic Scene Composition for Amodal Instance Segmentation
Deep Learning Course Project, Mar 2020 - May 2020
  • Developed a novel deep learning architecture, capable of generating realistic occluded compositions by placing provided individual objects in the scene, in a context-aware manner while preserving the original full masks of the occluded object.
  • Used a combined STN GAN framework to learn a projection matrix, based on Encoded Geometry and semantic information.
  • Evaluated the generated compositions by fine-tuning a pre-trained Mask-RCNN for the task of amodal instance segmentation and achieved a higher (~12%) COCO mean Average Precision as compared to a dataset with partial masks.
Semantic Scene Composition for Amodal Instance Segmentation more_vert
Deep Learning
Computer Vision
Research
Semantic Scene Composition for Amodal Instance Segmentationclose

Deep Learning Course Project, Mar 2020 - May 2020

  • Developed a novel deep learning architecture, capable of generating realistic occluded compositions by placing provided individual objects in the scene, in a context-aware manner while preserving the original full masks of the occluded object.
  • Used a combined STN GAN framework to learn a projection matrix, based on Encoded Geometry and semantic information.
  • Evaluated the generated compositions by fine-tuning a pre-trained Mask-RCNN for the task of amodal instance segmentation and achieved a higher (~12%) COCO mean Average Precision as compared to a dataset with partial masks.
CODE VIDEO DEMO REPORT
Penn Shell more_vert
Systems Programming
Operating Systems
Linux
C Language
Penn Shellclose

Operating Systems Project, Jan 2020

  • Implemented a fully functional shell that takes command from the user and executes them, in addition to functionalities like handling foreground and background processes, ,job control, standard input output file redirection, n- stage pipelines and asynchronous signal handling.
MORE ON GITHUB
Image Classification Caltech UCSD Birds 200 more_vert
Machine Learning
Deep Learning
Computer Vision

Image Classification Caltech UCSD Birds 200close

Deep Learning Homework MiniProject, Feb 2020

  • Trained and compared various classifiers such as Logistic Regression Classifier, Feed Forward Neural Network and Convolutional Neural Network to classify images of UCSD-Caltech Birds 200 dataset.
  • Improved accuracy by 20% by using transfer learning i.e. fine tuning pretrained ResNet50 despite scarce training samples.
CODE REPORT
Pixagram - Photo Sharing Social Network App more_vert
Software Development
Databases
Web
Node
Angular

Pixagram - Photo Sharing Social Network App close

Programming for the Web Course Project, Nov 2019 - Dec 2019

  • Developed photo-sharing social networking website that supports user authentication, posting photos, following users, liking photos, commenting on photos, editing posts and comments, and user recommendation system.
  • Developed backend services with Node.js and ExpressWe.js based on RESTful API, adopting Angular and Angular Material on client-side for UI and MySQL for storing information, along with unit and functional testing and deployed as Heroku app.
  • Used industry best practices in software development such as use case modelling, extreme programming, test driven development, documentation, continuous integration and, unit and functional testing.
CODE VISIT WEBSITE PROJECT WIKI
Efficient Duplicate Review Detection using Local Sensitive Hashing Techniques more_vert
Data Mining
NLP
Hashing

Efficient Duplicate Review Detection using Local Sensitive Hashing Techniquesclose

Data Mining Homework MiniProject, Feb 2020

  • Reduced time Complexity from O(n2) to O(n) of finding similar pairs of reviews and finding all the similar reviews given a query text efficiently using hashing techniques such as min hashing and local sensitive hashing on a dataset of about 160,000 user reviews from Amazon.
CODE REPORT
Yourooms - Meeting Room Booking App more_vert
Software Development
Databases
Web
PHP

Yourooms - Meeting Room Booking App close

Hobby Project, Nov 2017 - Dec 2017

  • Designed and developed a web application to facilitate booking of meeting and study rooms in the college, using HTML5, CSS3, JavaScript with PHP for connecting to the SQL database and storing booking information, time slots and services available within each room.
VIEW CODE ON GIHTUB
Yelp Restaurant Review Rating and Sentiment Prediction more_vert
Machine Learning
Deep Learning
NLP
Data Mining

Yelp Restaurant Review Rating and Sentiment Predictionclose

Machine Learning Course Project, Oct 2019 - Dec 2019

  • Performed data extraction and analysis of Yelp Restaurant reviews and built several baseline models such as Naïve Bayes, Decision Tree, Random Forest to predict ratings (out of 5) and sentiment (binary) of about 1M textual reviews.
  • Improved classification accuracies and F1 scores for multiclass classification by utilizing deep architectures like LSTMs and CNN along with weighted loss function and SMOTE (to handle class imbalance and majority bias).
  • Enhanced binary classification performance by building same models based on aspect descriptors instead of whole review.
CODE REPORT
Efficient Classification Techniques, Optimizations and Analysis on Fashion MNIST Datasetmore_vert
Data Mining
Machine Learning
Deep Learning
Optimization
Efficient Classification Techniques, Optimizations and Analysis on Fashion MNIST Datasetclose

Data Mining MiniProject, Mar 2020

  • Used optimization techniques such as Pegasos and Adagrad to train a linear SVM binary classifier in an online manner and proposed an algorithm to improve performance where Adagrad doesn't perform well.
  • Implemented a multiclassifier by using one vs one approach using the best trained SVM model from binary classifier.
  • Used a Convolutional Neural Network for multiclass classification in PyTorch, and made a comparative study of all the methods used.
REPORT AND ANALYSISCODE
Movie Recommender System based on Multi Armed Banditsmore_vert
Data Mining
Machine Learning
Recommender Systems
Optimization
Movie Recommender System based on Multi Armed Banditsclose

Data Mining MiniProject, May 2020

  • Implemented partial feedback and full feedback versions of stochastic and non stochastic approaches such as Epsilon Greedy, UCB1, EXP3, Thompson Sampling and multiplicative weight update algorithm to solve multi armed bandits’ problem, and modelled it to recommend movies to a user in an online manner with sublinear regret, given data about 1005 movies for 15000 days and performed comparative study to find which algorithm works best in different kind of situations.
REPORT AND ANALYSIS CODE
Clustering using Online K-means and K-means++ for IMDb datasetmore_vert
Data Mining
Machine Learning
Clustering
Optimization
Clustering using Online K-means and K-means++ for IMDb datasetclose

Data Mining MiniProject, May 2020

  • Performed feature engineering followed by clustering of movies in IMDb Title Dataset to cluster similar movies together.
  • Formalized the clustering problem as an optimization problem and used minibatch gradient descent to update the centers, and learnt them in an online manner, thereby improving performance than the usually used Lloyd's Algorithm.
  • Incorporated adagrad like learning rate for different centroids, based on the centroid being the nearest centroid to the points within the mini batch, therfore ensuring faster convergence.
  • Further improved the algorithm by using K Means++ initialization technique, based on picking successive centroids from a frequency distribution, proportional to the square of the distance between datapoints and the nearest centroid, which increases convergence rates and gives a better spread of centroids.
REPORT AND ANALYSIS CODE
Canny Edge Detection from Scratch more_vert
Computer Vision
Edge Detection
Python

Canny Edge Detection from Scratchclose

Computer Vision MiniProject, Sep 2019

  • Implemented Canny edge detector for images from scratch in Python without the use of external libraries like OpenCV.
  • Improvised Non Maximal Supression step by parallelizing code using meshgrid to calculate maximum gradient magnitudes of neighbouring pixels.
  • Performed Edge linking using discretized angles approach where angles were discretized into eight bins and also non discretized approach where pixels along edge were interpolated and edgemap was updated accordingly.
VIEW CODE AND RESULTS
Gradient Based Image blending more_vert
Computer Vision
Python
Computational Photography

Gradient Based Image blendingclose

Computer Vision MiniProject, Oct 2019

  • Implemented Gradient based Image Blending to seamlessly blend an object from a source image into a target image.
VIEW CODE AND RESULTS
Sequential Models for Text Classification and Generationmore_vert
Deep Learning
Natural Language Processing
PyTorch

Sequential Models for Text Classification and Generationclose

Deep Learning MiniProject, March 2020

  • Worked on problems of text classification to predict ratings and summary generation for Amazon reviews by using various approaches.
  • Preprocessed data and implemented simple baseline methods for the classification tasks, further used various sequential models such as GRU, RNN, LSTM and BiLSTM to implement the classification.
  • Improved the models by Incorporating additive self attention within the models to remember context and improve performance metrics.
  • Also, used bidirectional transformer based architectures such as BERT and RoBERTa based on HuggingFace to further enhance F1 score accuracy.
  • Designed seq2seq architecture with attention trained using teacher forcing strategy to generate summaries of review text.
VIEW CODE AND WALKTHROUGH
Deep Generative Models for Image Reconstructionmore_vert
Deep Learning
Computer Vision
GAN
Autoencoder

Deep Generative Models for Image Reconstructionclose

Deep Learning MiniProject, Feb 2020

  • Implemented a convolutional Autoencoder, and a Variational Convolutional Autoencoder to reconstruct images of shoes from a learned distribution on the UTZappos50k dataset and better understand Generative Methods.
  • Implemented a GAN and a LS-GAN to construct new instances of images from the shoes dataset and understand the intricacies and problems encountered while training a GAN.
VIEW CODE AND WALKTHROUGH
Object Tracking in Videos based on Optical Flowmore_vert
Computer Vision
Python
Computational Photography

Object Tracking in Videos based on Optical Flowclose

Computer Vision MiniProject, Nov 2019

  • Built an optical flow bounding box model to track the movement of multiple objects between consecutive frames of a video.
  • Used Shi-Tomasi Corner Detection for feature extraction and implemented Lucas-Kanade optical flow equation using Taylor series approximation to parallelly find the coordinates of all the features in successive frames.
  • Further computed the homography matrix based on the features and their positions in successive frames and used it to maintain the bounding box around the object being tracked.
VIEW RESULT

The website is a work in progress and still in development.

GET IN TOUCH TO KNOW MORE

Experiences

Graduate Research Intern
World Well Being Project, University of Pennsylvania
May 2021 - Present
  • Working on building a generalizable emotion based lexicon to get word level ratings by inverting and interpreting Deep Learning Models by techniques such as SHAP for text classification on textual data from social media platforms.
Graduate Teaching Assistant
University of Pennsylvania
Oct 2019 - Present
  • Guiding students, and actively participating in course content development, holding recitations, office hours, solving students' doubts and grading their projects and assignments for the following:
  • CIS520 Machine Learning : Class of 120 students taught by Prof. Lyle Ungar focusing on mathematical and application based foundations of machine learning.
  • CIS581 Computer Vision : Class of 40 students taught by Prof. Jianbo Shi focusing on image processing concepts and deep learning for vision.
  • Penn Fife CS Academy : Coding Instructor for High School Students of The Philadelphia School as a part of the Penn Fife CS Academy Program. Introduced the students to algorithmic thinking, and basic applications of Computer Science in gaming and robotics.
Volunteer, Volunteer Head
Child Rights and You, Pune PAG
Feb 2018 - Aug 2019
  • Involved in teaching students of two government-run schools in Pune, and conducting various activities and workshops for the welfare of the students.
  • Also, involved in planning and organising these events, managing and holding out inductions for recruiting other volunteers.
Summer Project Intern
Centre for Development of Advanced Computing, IN
Apr 2018 - Jun 2018
  • Worked on research, implementation, and development of an automated FAQ system (Chatbot) for assisting the applicants of CASB (Central Airmen Selection Board) Examination for entry into the Indian AirForce.
  • Researched on NLP applications related to Chatbots and implemented a generative chatbot based on a sequence to sequence LSTM with attention mechanism using Tensorflow (Python).
  • Built the production version of the bot using IBM Watson Assistant microservice, with a JSP wrap around deployed on an Apache Server for integration with the webpage.
Machine Learning Engineer Intern
Anomaly Solutions
Feb 2018 - Apr 2018
  • Worked on developing ideas and prototypes for the upcoming venture - Farmer's Eye aimed at assisting farmers in better agricultural practices.
  • Researched applications of machine learning, deep learning, and robotics that could be used to assist farmers in easing out their everyday schedules by automation and information.
  • Developed a model based on Convolutional Neural Networks and transfer learning to be able to classify images of plant leaves as healthy or diseased which could be deployed on low power hardware such as Raspberry Pi.
Public Relations Officer & Marketing Head
PICT IEEE Student Branch
Jul 2017 - Jul 2018
  • Involved in planning various activities, decision making and managing the various teams and sections within PICT IEEE Student Branch as a part of the Core Committee headed by Professor R.B. Ingle.
  • Involved in managing industrial relations and sponsorship management for the annual technical symposium Credenz' 17.
Software Development Engineer Intern
Airstacks Networks
Jul 2017 - Sep 2017
  • Developed front end modules in Angular, and RESTful APIs in PHP for handling the backend for their upcoming website.
Senior Web Developer
PICT IEEE Student Branch
Jul 2016 - Jul 2017
  • Developed web applications for and hosted the events 'Clash', 'Reverse Coding' and 'NCC (National Coding Contest)' for about 2700+ participants during Credenz'16 and Credenz Tech Days' 16.
  • The applications were to host online MCQs (Multiple Choice Questions) to test 'C' programming skills and competitive style coding competitions for the above-mentioned events.
  • The applications were based on the Django framework, used SQLite for database, and HTML5, CSS3, Bootstrap and MaterialiseCSS for the front end.

About Me

Hi, I'm Tejas Srivastava, a current master's student at The University of Pennsylvania, pursuing Computer and Information Science, working on a variety of applications of Deep Learning and Data Science from amodal segmentation in Computer Vision to explainability of Deep NLP Models.
I am currently working as a Data Science Research Assistant at the World Well Being Project, under the guidance of Professor Lyle Ungar, wherein I am engineering methods to build an emotion based lexica from Social Media data.
I believe that sharing ideas and teaching others is the best way to learn, and hence had appointments as Graduate Teaching Assistant for the course Computer Vision and Computational Photography taught by Professor Jianbo Shi, and Coding Instructor At Fife Penn CS Academy.

In 2019, I completed my Bachelor's in Computer Science Engineering from Pune Institute of Computer Technology at the University of Pune, India where I was also heavily involved in the functioning of PICT IEEE Student Branch as the Public Relations Officer, Marketing Head and part of the Technical Team. I have been a strong proponent of Artificial Intelligence and the positive impact it is making in businesses and personal lives alike, and have pursued various internships at organizations and startups like Centre for Development of Advanced Computing, India and Anomaly Solutions to explore various domains related to AI.
Back in India, I had also been involved with CRY, an NGO where I was leading the Pune Public Action Group along with a team of 30 volunteers, wherein we were teaching underprivileged students at a government-run school, and were involved in conducting awareness drives and workshops.
I believe in continuous learning and love learning about new things, problems and how technology is being employed to solve these problems. I also like to keep track of the latest advancements in technology, and am always excited to discuss latest products and feature releases. I like being motivated and keep chasing goals, and believe in self improvement, and learning from others.