Graduate Research Assistant
2018 - 2019 (over 1 year)
• Worked on Anomaly Detection, Urban Analytics Platform using Deep Learning. Currently exploring the use of Gating Networks in Meta Learning Shared Hierarchies for Reinforcement...more Learning. • Implemented a client-server model for customized 1-card Poker game for training RL Agent and deployed on Heroku. • Mentored and guided graduate students in completing their projects & assignments. Additional work included answering doubts & questions of students related to the subject.
•Implemented a new algorithm to improve the attribution of snippets in financial news article using NER, POS Tagging and Coref resolution. •Implemented a heuristic based...more module to extract the public companies relevant to a news article improving the extraction of companies by a factor of 50%. •Optimized and improved performance of deduplication module for new articles by factor of 10. •Forecasted the stock price using exponential smoothing of discrete time-series data using the stock news data.
A system to detect the drowsiness level in the drivers.
Deep Learning Engineer · • Trained Resnet-50 to classify the video using per-frame majority voting. Added LSTM layers to capture the temporal features… · More to improve accuracy to 95%. • Improved robustness of classification by implementing and training DCGAN.
Machine Learning Developer · Built a job recommendation system, which will compare the User's resume with Job Description and User's needs.… · More
• Predicted salaries by implementing Machine Learning models such as Naïve Bayes, Logistic Regression and K- means in Spark and clustered different types of jobs. • Built content-based recommendation system which recommended similar jobs based on user’s resume and job description.
Machine Learning Engineer, Python, Keras, TensorFlow · Used Deep Reinforcement Learning agent as a proxy for Generative Network in GANs to train a… · More robust neural network for classification.
• Implemented a novel way to use Deep Reinforcement Learning Agent as proxy for Generative Networks in GANs to train robust networks for classification. • Coded various RL algorithms like DQN, A3C to train DRL Agent.
Developed end to end Deep Learning model to predict traffic
Machine Learning Engineer, Python, Keras, TensorFlow · Developed an end-to-end system for ingesting multi- ple data sources and processing them via… · More deep neural network models (LSTM) to tackle predictive urban com- puting problems, like traffic flow predictions and modeling.
• Developed an end-to-end system for ingesting multiple data sources and processing them using deep neural network models (LSTM) to predict ETA of traffic flow with RMSE of 4.01. • Coded a web scraper to collect data continuously from Twitter, Waze, OpenWeather and Charlotte City. Distributed and synchronized scrapping and collection process on multiple computers.
Python, Scikit-Learn, Seaborn · Built a model using semi-supervised ML algorithms to detect insurance fraud claims.
- Did Exploratory Data… · More Analysis and Data Visualization. – Prototyped and tested predictive models for finding fraud claims submitted using hierarchical clustering. – Designed new features to increase the accuracy of the clustering of fraud claims.
Python, Machine Learning, Topic Modeling · Designed and developed a ML model to classify surprising news from a news corpus based on personalized… · More preferences using Topic Modeling and Supervised Machine Learning algorithms.
• Built a recommender system to provide news suggestion (similar and surprising) to a user. • Trained SVM & Random Forest (Scikit-learn) to classify documents using Topic modelling and Word Embeddings.
Algorithm Developer, Android, Java · Implemented a general game-playing agent for two-player deterministic games, using (1) minimax with alpha-beta… · More pruning, and (2) minimax-cutoff (i.e., with cutoff test to replace terminal test and with evaluation function to replace utility function) with alpha-beta pruning.
Applied it to the planar 3*3 Tic-Tac-Toe game and extended it to the planar n * n (n > 3 and n is odd) Tic-Tac-Toe game. Has a friendly graphical user interface to allow a human user to play the game against the algorithm.