|
I just completed masters in the Data Science Institute at Columbia University. I am currently on the job market, looking for exciting and impactful work in Machine Learning and its applications. I have a strong background and interest in the field of Machine Learning, Deep Learning and its applications to Computer Vision, Natural Language Processing, and Statistics. I want to harness the power of technology to be able to positively impact a larger section of the society. I graduated with a major in Mathematics and Scientific Computing and a minor in Industrial and Management from Indian Institute of Technology Kanpur in 2016. As part of my undergraduate thesis, I worked with Prof. Harish Karnick and Prof. Amit Mitra on a font free optical character recognition system for Devanagri Script. During the Summer of 2015 I worked as a data science intern at Fuzzy Logix under the mentorship of Mr. Partha Sen (CEO and CVO). Here I worked on the development of ensemble models for web based in-database analytics. During the Summers of 2017, I was fortunate to work as a Machine Learning Researcher at Netflix (Cloud Media Systems team), where I lead two independent projects on Computer Vision and Language.
|
|
|
|
|
 
|
Worked on classifying overall perceived emotion for the video using the change in facial expression of persons in the video over time.
Worked on generating captions for events taking place in a video with context to what has happened in the neighboring events. |
 
|
Working on extracting lesser known relationships based on some well known relationships from the financial newspaper articles data. The main idea is to create a large knowledge graph for different relationships. |
 
|
Analyzed effects of ensemble models on predictive power of Decision Trees, Logistic Regression, Neural networks and Naive Bayes model. Observed significant improvements in accuracy (by 5-15%) for ensemble comprising decision trees ( Random Forest) and neural networks, but not much significant improvement for ensemble of logistic regression Developed stored procedures for in-database analytic solutions on Netezza and SQL server. Was Offered a full time employment opportunity |
 
|
Modeled complex customer relationships through Markov Chain models using Statistical Analysis System to predict revenue band of customer at 94% accuracy |
|
|
 |
Devised algorithms for (samyutakshars) conjunct characters identification and separation into individual characters. The proposed algorithm decreases Word Error Rate by 10-15\% than current state of art techniques Work is submitted to ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP) |
|
|
 |
Utilized Extreme Gradient Boosting for feature extraction from 7 GB of training data, comprising of about 4200 features. Using the chosen subset of features, Random Forest (MCC Score: 0.40) and Gradient Boosting (MCC Score: 0.41) emerged the best models, to predict whether a product on the production line will be defective or not |
 |
Fused the artistic style of an artwork with the content of an image by maximizing the correlation between the generated image and feature map of VGG-19 net layers used to extract style and content from image. Provided a modified implementation of the Neural Algorithm for Artistic Style transfer as proposed by Leon Gatys in the Paper Also used Markov Random Fields for generating image (local image texturing) which gave better visualization of the fusion. |
 |
Implemented and compared different object detection, feature extraction (HOG, AlexNet) and classification methods (RandomForest, KernelSVM, Logistic Regression) on a large amount of unlabeled video surveillance data collected at IIT Kanpur Entity recognition using Deep Region based Convolutional Neural Net, feature extraction by penultimate layer of AlexNet, classification by Random Forest yielded best performance (91.75% accuracy) and real time processing |
 |
Combined historical usage patterns with weather data to forecast bike rental demand in the Bikeshare program in Washington, D. C. Formulated random forests and exponential gradient boosting ensemble to predict bike rental demand with 88% accuracy |
 |
Looked towards Support Vector Machines and Nearest Neighbors as Classification methods, that can be kernelized. Also looked at different formulations of SVM’s, namely C-SVM and nu-SVM. Depicted the wide applicability and ease of Kernel SVMs through real-world problems like in face detection, handwritten character detection, spam/non-spam classification |
 |
Systematically studied the existing techniques and literature for dimensional reduction. Explored and compared across different techniques through comparisons on different type of datasets |
 |
Applied binomial logistics regression to classify patients having recurrent breast cancer and patients having non-recurrent disease with 72%. Predicted the time after which tumour onsets again in recurring cases using multiple linear regression model |
 |
Predicted daily foreign exchange rate of Indian Rupee with USD, Euro, Yen, Great Britain pound using linear model (ARIMA) and compared the result with random walk model. Successfully conducted that any other ARIMA model prediction cannot beat random walk model predictions of daily foreign exchange rates |
|
|
|
|