Khusaal Giri

1) Developed an approach for performing Transfer Learning on Text Data using CNN & graphs.
2) Used word2vec model to generate graphs from text data & applied kernel PCA for dimensionality reduction, along with CNN procedure to perform classification on a different dataset belonging to the same domain.
3) Tools/Technologies Used: Python, Tensorflow, Scikit-learn, Google Cloud Platform.

1) Trained a CNN network on the MNIST dataset which comprised of only sevens with an added additional special effect.
2) Used the output of the hidden layer as a latent embedding for the training dataset which acted as an encoder of VAE.
3) Posed the learned latent embedding as a Multivariate standard normal distribution & updated the loss function accordingly.
4) Randomly sampled code vectors from the learned Multivariate Gaussian distribution to generate new digits which acted as a… Mehr anzeigen

1) Trained a CNN network on the MNIST dataset which comprised of only sevens with an added additional special effect.
2) Used the output of the hidden layer as a latent embedding for the training dataset which acted as an encoder of VAE.
3) Posed the learned latent embedding as a Multivariate standard normal distribution & updated the loss function accordingly.
4) Randomly sampled code vectors from the learned Multivariate Gaussian distribution to generate new digits which acted as a decoder of VAE.
5) Used the new generated digits to verify the special effect added in the original training dataset.
Tools/Technologies used : Pyton,Tensorflow Weniger anzeigen

1) Trained a CNN network on MNIST dataset & stored all the weight matrices in numpy files.
2) Developed a new CNN architecture for the same dataset & used the singular value decomposed matrices of the stored weights in reduced dimension as the new weights of the network.
3) Compared the accuracy of the network which uses weights of varying dimension derived from the stored weights.
4) Tools/Technologies used : Python,Tensorflow

Speaker Verification using Siamese Network

1) Created the dataset for Speaker verification by forming pairs of speech signals & assigned the labels by posing it as a binary classification problem wherein positive class represents that the pair belongs to the same speaker & negative class represents that the pair belongs to different speakers.
2) Trained a Siamese network which provides a fixed dimension embedding to the speech signals.
3) Used the generated embeddings to compute… Mehr anzeigen

Speaker Verification using Siamese Network

1) Created the dataset for Speaker verification by forming pairs of speech signals & assigned the labels by posing it as a binary classification problem wherein positive class represents that the pair belongs to the same speaker & negative class represents that the pair belongs to different speakers.
2) Trained a Siamese network which provides a fixed dimension embedding to the speech signals.
3) Used the generated embeddings to compute the similarity between two speech signals thus assigning the class label based on the similarity score obtained for the pair.
4) Technologies Used : Python, Tensorflow Weniger anzeigen

1) Developed a method for Image segmentation to generate Individual alphabets from an image of handwritten words using Neural Network & Skeletonization.
2) Generated artificial data to produce the positive & negative examples for the segmentation task by adding some distortion to the training dataset & used it for training the Neural network.
3) Trained the IAM dataset for the individual alphabet classification & recognition in Scanned documents using Convolutional Neural Network… Mehr anzeigen

1) Developed a method for Image segmentation to generate Individual alphabets from an image of handwritten words using Neural Network & Skeletonization.
2) Generated artificial data to produce the positive & negative examples for the segmentation task by adding some distortion to the training dataset & used it for training the Neural network.
3) Trained the IAM dataset for the individual alphabet classification & recognition in Scanned documents using Convolutional Neural Network (CNN).
4) Tools Used: Python, Tensorflow, OpenCV. Weniger anzeigen

1) Built an efficient Factorization Machine (FM) to analyze the Netflix Movie Ratings Dataset; which comprises of about 18,000 movies & 5,00,000 users.
2) Addressed the issue of dealing with large Sparse matrices generated during the training procedure of Stochastic Gradient Descent (SGD) by having a distributed implementation of FM with Linear time complexity.
3) Tools Used: Python, PySpark, Hadoop, Matplotlib.

1) The CMS(Centre of Medicare & Medicaid services) data contains values for Average Medicare payment amount(Response variable),Discharge rate of the hospital,Disease category & the region where the hospital is located.
2) We examined the CMS data by plotting the relationship between the response variable & each predictor variable individually using ggplot2. This relationship also helped us in deciding whether there exists an interaction between the variables or not.
3) We also analysed… Mehr anzeigen

1) The CMS(Centre of Medicare & Medicaid services) data contains values for Average Medicare payment amount(Response variable),Discharge rate of the hospital,Disease category & the region where the hospital is located.
2) We examined the CMS data by plotting the relationship between the response variable & each predictor variable individually using ggplot2. This relationship also helped us in deciding whether there exists an interaction between the variables or not.
3) We also analysed the trend between the average Medicare payment amount & the other variables in our data by fitting a Generalized Linear Model(glm) on the sample in R programming language. Weniger anzeigen

1) Studied the OpenFace dataset & understood the different Action Units & landmark points of a face which are associated with every human emotion.
2) Trained the Openface dataset using Collaborative Filtering algorithm & used Time Series Analysis to detect the basic human emotions of the user while watching a video.
3) Generated the overall feedback of the user for the video being watched using Random Forest Classifier by predicting the objective answers to a pre-defined set of questions… Mehr anzeigen

1) Studied the OpenFace dataset & understood the different Action Units & landmark points of a face which are associated with every human emotion.
2) Trained the Openface dataset using Collaborative Filtering algorithm & used Time Series Analysis to detect the basic human emotions of the user while watching a video.
3) Generated the overall feedback of the user for the video being watched using Random Forest Classifier by predicting the objective answers to a pre-defined set of questions which are mainly focused on the likeliness of the video being watched. Weniger anzeigen

1) Implemented Multi-image panorama creation using C++ language and CImg and armadillo library.
2) Executed Image matching algorithm using SIFT Descriptors of the images.
3) Implemented RANSAC algorithm for computing the homographic matrix and the orientation between two images.
4) Applied Inverse warping and blending techniques for stitching two images together.

1) The CCES(Cooperative Congressional Election study) data contains values for Attitude towards Immigration,Education level,Gender,Party Affiliation & Race of the voters.
2) From the given data,we derived new variables which corresponds to the weighted proportion of each given variable for each demographic.
3) Then,we studied the effect of attitude towards immigration on the switching of votes of 2012 Obama supporters who became 2016 Trump supporters using a single,two & all predictor… Mehr anzeigen

1) The CCES(Cooperative Congressional Election study) data contains values for Attitude towards Immigration,Education level,Gender,Party Affiliation & Race of the voters.
2) From the given data,we derived new variables which corresponds to the weighted proportion of each given variable for each demographic.
3) Then,we studied the effect of attitude towards immigration on the switching of votes of 2012 Obama supporters who became 2016 Trump supporters using a single,two & all predictor variables with their calculated weights & interactions into consideration.
4) The data can be obtained here: https://cces.gov.harvard.edu/data Weniger anzeigen

1) Developed a Object Recognition module,which detects the presence of IC(Integrated Circuit) on PCB(Printed Circuit Board) Images using C++.
2) Implemented Sobel filtering and applied Hough Transform to an Image to detect the edges of the IC on PCB.Also,used Non-Maximal suppression & modified accumulator updation for higher precision and accuracy.

1) The data repository Gapminder contains values for life expectancy, GDP per capita, and population, every five years, from 1952 to 2007.
2) We analysed the data by breaking this time series down into 3 components(trend,seasonal & oscillation) for all the countries & finding out the trend in the life expectancy values.
3) We also studied the effect of GDP per capita on the life expectancy of different countries over time.
4) The data can be obtained here:… Mehr anzeigen

1) The data repository Gapminder contains values for life expectancy, GDP per capita, and population, every five years, from 1952 to 2007.
2) We analysed the data by breaking this time series down into 3 components(trend,seasonal & oscillation) for all the countries & finding out the trend in the life expectancy values.
3) We also studied the effect of GDP per capita on the life expectancy of different countries over time.
4) The data can be obtained here: https://cran.r-project.org/web/packages/gapminder/index.html Weniger anzeigen

1) Created a module for classifying the image orientation into 4 categories: North, South, East & West using Adaboost, Backpropagation & kNN algorithm in Python.
2) Worked on feature extraction of the training images based on the pixel color & position to correctly incorporate them into the learning algorithms.

1) Implemented Personalised Recommendation Engine for purchase of Grocery items,using clustering Algorithm in Python.
2) Implemented Recipe based Recommendation Engine,where recipes are recommended to the Customers using Market Basket Analysis.

1) Developed an Online Placement Management System for handling various Placement Activities of the College using Java, SpringMVC and Hibernate.
2) Worked on the Login Authentication & Authorisation module.