Hi! I'm Pradeep.
e-mail: pdeepsingh094@gmail.com

I'm passionate about data and science. I enjoy working on complex problems that requires me to distill technical research in related areas and adapt it to the problem at hand. My area of interest lies at the intersection of Machine Learning and Data Science in general and Deep Learning for Computer Vision/ NLP in specific.

I'm in my greatest element when working with any kind of visual data. In the past, my work has involved using data of multiple modalities - visual, (un) structured mesh, text - all of which are both exciting and challenging to work with. In my free time, I like reading and writing on Quora, cooking and hiking.

I completed my Masters in Computational Data Science at San Diego State University, supervised by Patrick Shoemaker. My thesis investigated neural mechanism(s) for target (object) tracking in Insect Visual System. In past, I did my undergraduate at University of Mumbai and have spent time at Dassault Systèmes, Raman Research Institute and HERE Technologies.

Accelrating CFD Simulations using Machine (Deep) Learning
Developed a novel machine learning framework (SRCFD) for accelerating CFD simulations by super-resolving coarse resolution simulation into fine resolution simulation using (graph) convolutional neural network.

MS Thesis: Models for Propagating Facilitation in the Insect Visual System.
Flying insect species like dragonflies are capable of predicting the path or location of their target even if the target has occluded by some object for some period of time. This ability to predict the path is supported by a processing mechanism which is called response facilitation. I'm working on modeling this processing mechanism: response facilitation in small-target- sensitive visual neurons in dragonflies.

Built and trained 5 different Convolutional Neural Networks using Keras and TensorFlow to classify 70,000 fashion images into 10 labels. Achieved accuracy of 95% with VGG model + batch normalization.

Built a end-to-end machine translation pipeline using recurrent neural network which takes English text & return it's French translation. Experimented with various models: simple RNN, RNN with Embedding, Bidirectional RNN, Encoder-Decoder RNN & achieved accuracy of 98%.

Autocoders are a family of neural network models that aims to learn compressed latent representation of high-dimensional data. In this project, I study, review and implement autoencoders in various forms: basic autoencoder, denoising, sparse, and contractive autoencoders, and then Variational Autoencoder (VAE) and its modification beta-VAE. The goal of this project is to study and understand how autoencoders (and it's variants) work.

Implemented SOTA Few-shot learning models like, Siamese neural network, Matching Networks and Prototypical Networks in TensorFlow.

Implemented SOTA Image super-resolution research papers – SRCNN, FSRCNN, ESPCN, SRGAN, EDSR and WDSR in TensorFlow. Explored approaches like adversial training, sub-pixel convolution.

This project explores Bayesian optimization techniques for hyperparameter tuning in machine learning algorithms and compare it with different methods like: manual search, grid search, random search. Goal of this project was twofold: 1) To study how bayesian optimization can be used in hyperparameter tuning in order to improve the current methods, and 2) Comprehensive analysis of hyperparameter optimization algorithms in Machine Learning.

Implemented a sequential and parallel neural network model using data based parallelism in Python using MPI and GPU computing. Achieved 50% speedup in the training time.

Predicting Customer churn rate on Telco Customer churn dataset, taken from kaggale. We'll take a look at what types of customer data we have, do some preliminary analysis, and develop churn prediction models - all with Python/PySpark and different machine learning frameworks, like, ML Package and Scikit-learn.

Built an end-to-end sentiment classification system using Recurrent neural network and Naive Bayes classifier to classify the sentiment of 50,000 movie reviews in IMDb dataset.

• Advice: How to Succeed in Graduate School
• Cheatsheet for Machine Learning, Deep Learning and Probablity
• Some (cool) courses: Deep Learning for NLP, Deep Learning for Visual Recognition, Data 8: Foundations of Data Science
• Blog's that I admire: Ferenc Huszar’s blog, Distill, Chris Olah’s blog, David Ha’s blog

"If we want machines to think, we need to teach them to see" ~ Fei-Fei Li
Template borrowed from Jon Barron.