Projects

Generative Model for Text - Machine Learning, Natural Language Processing (June - July 2020)

  • Built a generative machine learning model to mimic the writing of prominent author by making text corpus of 7 books, encoded to extended ASCII code and implemented LSTM model with choosing window size and hidden layers and predicting 1000 characters for a sample text with a loss of 1.658.


Image Classification using Successive Subspace Learning – Image Processing & Deep Learning (March - May 2020)

  • Modelled the LeNet5 architecture with Convolutional Neural Network and Successive Subspace Learning (SSL) varying the hyperparameters, optimizers to CIFAR10 dataset for image classification and achieving 82.54% testing accuracy.

  • Concepts used: Python, Tensorflow, Keras, Deep Learning, PixelHop, PixelHop++


Hand Postures (Motion Capture) Classification - Mathematical Pattern Recognition (February 2020 – May 2020)

  • Performed exploratory analysis to extract features from raw data of hand postures dataset then performing LDA for classification using classifiers like SVM, RF, KNN, Naïve Bayes, Perceptron and optimized performance by 95%.

  • Enhanced better understanding of results by iteratively revising and using cross validations to get desired outcome using Python’s Scikit Learn and fine-tuning classification methodologies resulting in 100% verified output.


Image Quality Enhancement, Defect Detection, Half Toning – Digital Image Processing (Sept. – Dec. 2019)

  • Implemented Bilinear Transformations and MHC methods to reconstruct full color image and ameliorated quality by 60% using denoising techniques and deployed warping algorithms improving image visualization by 65%.

  • Executed morphological operations to count stars, PCB holes and paths to facilitate defect detection by Connected Component Labelling (CCL). Extracted image features by half toning and increased image printing efficiency by 86%.

  • Technologies implemented: MATLAB, Denoising, Edge Detection, Image warping, CCL, Half toning algorithms.


Bigram Feature Vector Generation – Machine Learning

  • Scraped and extracted data from Web using HTTP GET request command and developed feature vector of their bigrams and computed word frequency count using Python dictionaries.

  • Converted bigrams strings into their hexadecimal and decimal equivalents for efficient storage of data.

  • Techniques used: Python, HTTPs requests, n-grams, UTF-8 encoding and Typecasting.


Gesture and Voice Control Mouse – Natural Language Processing (August 2018 - March 2019)

  • Led group of 4 engineers to develop hand gloves using NLP voice synthesizing in android and achieved 100% accuracy.

  • Involved Technologies: Arduino IDE, Building and Debugging circuits, Android App Development, Voice Control.


Real/ Fake Currency Detection (August 2018- November 2018)

  • Demonstrated Supervised Neural networks; Collected data to train NN.

  • Evaluated those to test unknown notes and processed with less computational time.