Audio Style Transfer
The motivation behind this project was to use the technique of neural style transfer to capture the style of one song and transfer it to the other.
The technique behind style transfer is to separate the content and style of a song using a pre-trained deep network and transfer the style alone to another song. You can listen to some interesting results below.
Artificial Intelligence Course, Carnegie Mellon
Question - Answering system based on Wikipedia Articles
This was a common project for the class. Each team had to code a question-answer system which generates intelligent questions from wikipedia articles and answer questions as well. The teams were pit against each other and our team finished in the top 5.
Spring 2017Natural Language Processing Course, Carnegie Mellon
Self-photo mosaic
We chose to learn about the growing topic of neural style transfer and implement a cool idea. So, we made an image learn its own features. Each tile in the image is the image itself!! But, after learning the features of that particular tile from a pre-trained network.
Computer Vision Course, Carnegie Mellon
Recurrent Network for Vision
Based on a paper released in 2016, our project aimed to recreate the effect of a recurrent network for visual classification. The introduction of recurring layers in a deep neural network based on a shared weight learning reduces the complexity of the network while maintaining similar levels of accuracy.
Pattern Recognition Course, Carnegie Mellon
Food recommendation system
A recommendation system based on just reviews is very common. We took it a step further to implement colloborative filtering and sentiment analysis to find out the best and worst dishes at each restaurant based on reviews from zomato.com
Practical Data Science Course, Carnegie Mellon
Hyperspectral image classification using tensors
This was the stepping stone to my career in Machine Learning at CMU. My under-graduate thesis project focused on exploiting the tensorial features of Hyperspectral images. This is where I really learned about Dimensionality reduction, Image segmentation, Multi class classification. The project was presented in WISPNET 2016, IEEE conference and published on IEEEXplore. etc.
Thesis Project, SSN College of Engineering