Name: Satyaki Sanyal
Type: User
Company: @americanexpress
Bio: An avid Kaggler, a coder and a foodie. By passion and profession. I am a final year engineering student looking for a job in R&D with Artificial Intelligence.
Blog: http://www.satyakisanyal.me/
Satyaki Sanyal's Projects
TThis project is a webapp which shows credit card statements and graphs the data. This is the solution of a challenge given by American Express.
This project is build up completely with numpy. It implements basic neural network concepts including backpropagation, hidden layers, activation function and gradient descent.
This project uses Mini-learn on Boston's housing data-set. Mini-learn is a miniature version of tensor-flow which I made to play around with neural nets. See https://github.com/Satyaki0924/minilearn for more information.
This project uses convolutional neural network to recognise digits from images. This project has been trained on MNIST dataset. This project will soon be updated to recognise custom images.
This project uses a sequence to sequence model of the recurrent neural network to translate any piece of English text to French. I have used recurrent nets because while training on huge data, recurrent nets actually predict the outcome a lot better than any normal machine learning models. In this specific model, the data first passes through an encoder, comes out as an understanding and passes to a decoder. The decoder generates the output.
Mini-learn is a miniature version of tensor-flow which I made with ONLY NUMPY to play with perceptrons. You can use this project like you use tflearn. Go to https://github.com/Satyaki0924/boston-housing-with-minilearn to see it's usage.
This project analyses sentiments using deep learning. I have used only numpy to make the neural network. It portrays basic deep learning features using feed forward, backpropagation, gradient descent and activation functions.
This project analyses live twitter sentiments and visualises them using recurrent neural networks and long short term memories.
This project generates TV scripts with Recurrent Neural Networks and LSTMs. This project is trained on a script of the famous American sitcom, The Simpsons.