Some of the machine learning(and deep learning) competitive and non-competitive implementaions.
Neural Machine Translation
May 2019 - Jun 2019
Machine Translation using LSTM with a seq2seq architecture for translation from English sentences to Hindi sentences.
Toxic Comment Classification
Apr 2019 - May 2019
Classifying toxic comments from the internet into 6 classes as multilabels and calculating the probablity for each class. It's an old Kaggle competition problem. Achieved accuracy of 98% appx on dev set . While accuracy on Kaggle was 91.2%.
Poetry Generation
Apr 2019 - Present
Generating poetry using encoder and decoder RNN model(seq2seq).
News Article Summarization
Jun 2019 - Present
Using abstractive and extractive methods to summarize a given news article in four to five sentences by applying machine learning and deep learning techniques.
Hand-Written-Digit Recognition
Mar 2019 - Present
Using Deep Neural Network to recognize the hand written digits. The data is a part of the MNIST data and was available on Kaggle as a part of competition.
Accuracy obtained on Kaggle = 97.328
Accuracy on own test set = 98.8
Classification of text files and extracting specific features from them.
Mar 2019 - Present
This is based on an assignment from DBS Partners. Using Natural Language Processing to classify text files into two classes .Furthermore extracting specified features from those text files . So, far part 1 is completed with accuracy of 90.01%.