BIDIT DAS

**** ******** **., ***#***, Houston, Texas-77025 832-***-**** adg698@r.postjobfree.com https://github.com/biditdas18 https://www.linkedin.com/in/biditdas/

Experience

Deep Learning Researcher, University of Houston June 2019 – Present

Collaborated with a group of researchers in the Network and Cybersecurity lab for research and development of New methods for Intrusion detection in Networks and prevention of most common attacks such as Denial of Service (DoS), Worms, Fuzzers, Exploit and Analysis attacks.

Successfully Developed machine learning algorithms such as Logistic regression, Random Forest, SVM, Gradient Boosting for detection of Network attack in the UNSW_NB15 dataset which obtained the highest accuracy of 97%.

Created a deep Artificial Neural Network of 20 layers to specifically detect DoS attack in a large unbalanced dataset which obtained an accuracy up to 90% after feature selection demonstrating the superiority of this method over methods designed specifically for DoS.

Associate Systems Engineer, IBM GBS, India February 2018- May 2018

Assisted managers in understanding the business requirement and fetching relevant data from database using MySQL to prepare reports.

Collaborated with a team of software engineers for management and generation of sales report for various clients using JD Edwards.

Projects

Face Detection and Recognition System

The goal of this project was to develop a face detection and recognition system that has become popular in recent years and is highly trusted as a security identification feature in smart devices. The objective was to check the accuracy and suggest improvement in this system

Built a system that would localize faces in images of the custom dataset and perform preprocessing steps like face alignment and cropping.

Developed a neural Network that would take the output of the preprocessing steps as input and create a 128-d embedding that would quantify the face itself using a triplet loss function.

Finally developed a linear SVM model to operate on the top of 128-d embeddings and perform classification and recognition of known faces in the image obtaining a confidence level as high as 70%.

Generative Adversarial Networks for Generating New Faces

This project was developed to explore the capabilities of Generative Adversarial Network which is the only kind of a Deep Learning Model that has the capability of Generating new data instead of operating on existing ones.

Created a System consisting of two CNNs joined in conjugation where one network was a generator responsible for generating new faces from the latent space and the other network was a discriminator responsible for discriminating between real and generated images.

Designed the conjugate system to first train the Discriminator network to discriminate images generated by the generator, then trained the combined model of generator and discriminator where weights of discriminator were fixed.

The model improves over each epoch and was ultimately successful in generating new faces almost 60% of the time in each epoch.

Covid-19 Detection from Chest X-Ray Images

This project was inspired by the recent events of the Covid-19 pandemic and is designed to show the potential power of deep learning model for Detecting Covid-19 from X-ray Image and to encourage research for developing a medically approved version of this method.

Developed a Convolutional Neural network VGG-16 model with the weights from Imagenet dataset along with a custom dense layer and SoftMax layer to detect Covid-19 infection from chest X-Ray.

Designed the network to freeze the Imagenet weights obtained through transfer learning and train the custom dense layer and SoftMax layer over a custom dataset of normal and Covid-19 X-ray images to obtain a detection accuracy of 98.81%.

Automated Image Captioning System

This project was developed to recreate the conditions of the most recent innovation of automated image captioning in the deep learning community and to study the scope and application of this technology.

Designed a Convolutional Neural Network Resnet model with the last layer removed and an LSTM model, to convert an image into a feature vector and to covert processed text samples into word vectors respectively.

Created a combined model with LSTM and Resnet that would take in the image vectors and the word embeddings and would generate the next word based on the highest probability of predicted words until the caption is complete or the caption length limit is reached.

The resultant model obtained a high degree of accuracy in caption generation based on images.

Skills

Tools and Libraries: Python, C++, MATLAB, TensorFlow, PyTorch, Keras, Pandas, Numpy, Matplotlib, Scikit-Learn, OpenCV, MySQL, Tableau

Techniques Experienced: Image Classification, Deep Neural Network (DNN), Convolutional Neural Network (CNN), Recurrent Neural Networks (RNN), Natural Language Processing (NLP), Generative Adversarial Networks (GAN), Clustering Algorithms, Regression Techniques, Classification Techniques, Exploratory Data Analysis, Hyperparameter optimization, and tuning, segmentation, object detection, tracking

Education

M.S., Electrical and Computer Engineering (GPA: 3.7/4.0) August 2018- May 2020

University of Houston, Cullen College of Engineering, Houston, TX

Coursework: Machine Learning, Deep Learning and Computer Vision, Digital Image Processing, Stochastic Process

B.Eng., Electrical Engineering (GPA: 3.5/4.0) August 2013 - June 2017

KIIT University, Bhubaneswar, Odisha, India

Contact this candidate