Praphul Singh

I am working as a Deep Learning Engineer at Oracle, Bangalore. My interest lies collectively in Natural Language Processing, Domain Adaptation, Deep Reinforcement Learning, Graph Neural Networks and Capsule Networks.

I graduated from Indian Institute of Technology, Kanpur with a major in Electrical Engineering in 2019. I took a variety of courses on NLP, Visual Recognition, Data Mining, Neural Networks, Image Processing to strengthen my grasp in the field of machine learning.

At Oracle, I have been working with the SR Analysis team on Language Understanding Models using highly imbalanced datasets exploring various techniques based on LSTMs, Multihead Attention, Transformers, and Graph Convolutional Neural Networks. I also have a Reinforcement Learning Series published on Oracle Datascience Blog.

I have some patents and research papers submissions in the process as well, which I intend to mention once they get approved.

Email / LinkedIn / Resume / Google Scholar / Personal Blog

Paper Implementations
	Text-Based Graph Convolutional Networks Explained the basics of graph neural networks followed by graph convolutional networks and using it for language understanding.Also, implemented a research paper based on gcn in Tensorflow Keras.
	Capsule Networks and Dynamic Routing between capsules Explained the basics of capsule networks followed by the dynamic routing mechanism between the capsules and implemented a capsule network in Pytorch for image classification.
	Gradient-weighted Class Activation Mapping Explained and implemented the grad-cam in Pytorch for finding the class activation mappings for a given (image,class) pair.
	Incremental Learning Without Forgetting Tensorflow Keras implementation of a research paper based on continual learning which tries to solve the problem of catastrophic forgetting by pseudo labelling the new data.
	Generating Handwritten Sequences Using LSTMs and Mixed Density Networks Implemented and succesfully trained a model in Tensorflow based on the research paper by DeepMind which uses LSTM layers followed by Mixed Density Networks to generate handwriting strokes.
	BiDirectional Attention Flow Model for Machine Comprehension Implemented the research paper in Tensorflow which explains a mechanism to achieve fine representation of context and query by learning a common similarity matrix and a bidirectional attention mechanism i.e., Context2Query and Query2Context.
	Multi Head Self Attention Studied and explained the research paper, Attention is all you need which proposes a new way to attending to the encoded information by the means of multiple heads.
	Breaking into Transformers Explained and implemented the fundamental blocks of a bert transformer in Tensorflow Keras.

inspired from this website