Academic
1. Hand written devanagari script recognition
The project aims to build machine learning models that can accurately identify handwritten characters in Devanagari script. It provides pre-processed datasets, training scripts, and evaluation tools for researchers and developers interested in character recognition.
Bachelors project 2015 GEC Goa.
2. Total Electron Content(TEC) Data as Seismic precursor using Multi Fractal Detranded Fluctuation Analysis(MFDFA) "Total Electron Content (TEC) data acts as a seismic precursor by utilizing Multi Fractal Detrended Fluctuation Analysis (MFDFA). MFDFA helps in identifying seismic activities by analyzing fractal patterns and fluctuations in TEC data, enabling early detection of earthquakes through ionospheric variations."
Masters project 2017 GEC Goa & IIT Bombay.
3. Hybrid Evolutionary Optimization(HEO) of Generative Adversarial Network(GAN)
Hybrid Evolutionary Optimization (HEO) of Generative Adversarial Network (GAN)¶
Hybrid Evolutionary Optimization (HEO) integrates evolutionary algorithms with Generative Adversarial Networks (GANs) to enhance the training process. HEO optimizes the GANs’ parameters by combining genetic algorithms and backpropagation, leading to faster convergence and improved performance in generating high-quality images. This hybrid approach leverages the strengths of both evolutionary strategies and deep learning.
Hybrid GAN with MFDFA at Discriminator to Generate Faster Images¶
A Hybrid GAN employing Multi-Fractal Detrended Fluctuation Analysis (MFDFA) within its discriminator accelerates image generation. By analyzing fractal patterns in input data, MFDFA enhances the discriminator’s ability to distinguish between real and generated images. This integration results in a more efficient training process, reducing computational time and producing images more quickly.
2D MFDFA Replacing Discriminator in GAN¶
Replacing the traditional discriminator in a GAN with a 2D version of Multi-Fractal Detrended Fluctuation Analysis (MFDFA) introduces a novel approach to image generation. 2D MFDFA evaluates complex temporal and spatial fractal properties of input data, refining the GAN’s learning process. This substitution enhances the network’s ability to generate realistic images by leveraging multifractal analysis.
PhD 2019* NIT Goa.