Projects
Please find below some of the interesting projects that I have worked on!
Improving Simple VSM Search Engine
Description:
​
Analyzed the limitations of the search engine implemented previously using a simple Vector Space Model (VSM) TF IDF Score-based document Vectors and implemented a better retrieval model on the Cranfield Dataset.
Click here to view the video presentation.
​
Technologies:
Python and Natural Language Processing.
Train a CNN model from scratch
Description:
​
Trained a CNN model from scratch and learnt how to tune the hyperparameters and visualise filters, finetuned a pre-trained model and used YoloV3 model to build a Real Time Wildlife Educator for Children.
​
Technologies:
Python, Convolutional Neural Networks using TensorFlow and wandb for visualization.
Creating a Transliteration System using RNN
Description:
​
Implemented a transliteration system using RNN, LSTM, GRU on Dakshina Dataset, overcame limitations of vanilla sequence to sequence model using attention networks and visualized the interaction between different components in the input for producing the output in multiple ways.
​
​
Technologies:
Python, HTML, wandb and Deep learning.
Neural Network Model Development from scratch
Description:
​
Implemented a feed-forward neural network and used gradient descent (and its variants: momentum, NAG, RMSProp, ADAM, NADAM) with backpropagation for classifying images from the fashion-MNIST data over 10 class labels.
​
​
Technologies:
Python, wandb and Deep learning.
by Arup Das