My Research Work and Projects

Here are some of the projects that I have done along with the research experience I have. They are a mixture of course projects, technical projects and self-learning projects. I will provide detailed info about the projects that I am currently working on after they are finished.

Research

• Learning to Speak on Behalf of a Group: Medium Access Control for Sending a Shared Message
  The project involved developing an algorithm for multiple agents to learn Medium Access Control (MAC) protocol to communicate through a set of collision channels. Our problem required agents to share information without collision at least once at any instant, which is different from the conventional average throughput maximization. We modeled our problem by posing it as a Multi-Armed Bandit for each agent where it learned successfully which channels to access when it became active. By simulating over multiple cases, we showed the superiority of our learning-based solution over heuristic approaches that utilize the knowledge of the distribution according to which observer agents are chosen. We theoretically proved the NP-hardness of the problem and also showed that it is sufficient to search for the optimal solution only in the space of deterministic actions. We have submitted our work as a conference paper at IEEE WCNC 2022. For more information please refer to my CV.

• Micro-Doppler Effects in RADAR
  I was involved in this research during my sophomore year. In real life, we have sources like helicopters that have vibrating surface (due to engine) and rotating blades. These motions induce a sinusoidally varying frequency in the RADAR signal along with the body Doppler. If we can estimate these frequencies we can even tell the frequency of vibration of the engine and the angular speed of the blades. Our task is, given a signal, we need to estimate the values of the micro-doppler and body-doppler frequencies. For more information please refer to my CV.

Projects

• Deep Learning for Channel Coding via Neural Mutual Information Estimation
  This was my course project for Machine Learning course. We implemented the idea of auto-encoding message signals by maximizing mutual information between input and output of the channel. The fascinating thing about our project was that without knowing the probability distribution of the channel, the model encoded the signal and produced accuracy similar to QAM, the best method to encode messages. For more information please refer to the documentation
• Statistical Compressed Sensing of Gaussian Mixture Models
  In this project, we explored reconstruction of images without any prior knowledge of the probability distribution parameters. Compressed sensing deals with the domain where even with less number of linear measurements of the signal values we can accurately reconstruct the signal. Using EM-algorithm to estimate the distibution parameters and then reconstructing the image using MAP estimator produced even better results than conventional compressed sening. For more information please refer to the documentation
• Fischer Faces vs Eigen Faces
  This was a course project which I did with two other team members. Fischer Linear Discriminant is a classification algorithm which tries to project data into a dimension where within class scatter is minimized while between class scatter is maximized. We implemented this algorithm on CMU Facedataset and Yale A and Yale B face dataset and found the results to be fascinating. The algorithm did very well with Yale dataset but performed very poor on the CMU dataset. For reasons please refer to the documentation. As an extension, we also used FLD to classify images with people having glasses or without it.
• Temperature Display and Controller
  This was full semester course project where we had to design a temperature controller which could be easily interfaced with a day to fay electrical appliance such as AC or refrigirator. Apart from that we also design a circuit to display the current temperature of the environment on LCD screen. For display, we used PT-51 microcontroller and for temperature control we designed a simple ON-OFF circuit. For complete schemetic and more information please refer to the documentation.
• Autonomous Security Bot
  We developed a bot that could be used in residential or industrial complexes. It not only identifies strangers but also guides people, like guests or visitors who come to the complex first time, to the right destination. Our idea was to use a microprocessor(Raspberry Pi in our case) which would store the map of the complex and data of all the residents in it. When it encountered a person it would straightaway mail the security guard of a potential security breach and if the person was a credible visitor it would guide him/her to the destination. For complete information please refer to the documentation.
• Image Inpainting Via Sparse Representation
  This was a self-motivating project I did during my holidays. I had read about the sparse representation of signals using Dictionary Learning. I searched more about this idea and came across a paper which had employed this technique to Image Inpainting. It is the process of hiding an object in a target patch in an image in such a way that when someone looks at the image it appears to him the object never existed in the image. Please refer to this link for my full implementation.
• Visualization in Deep Learning using Grad-Cam
  This was also a self-motivating project to check how much I had learned in Deep Learning. One of the major work in Deep Learning is to study how does the filter/kernals work. Grad-Cam is a technique to study which part of an input image our network is focussing on. For full implementation please refer here.
• Smiley Face on 8*8 LED Matrix
  This was a course project which I did when we were first taught about digital circuits in electronics. Our task was to generate a Smiley Face on 8 by 8 LED matrix using digital circuits. The idea was that at a specific time only one row of the matrix would be lit and this process would be repeated for each row so fast that to our eye it would feel like the whole figure is generated at once. For more information please refer to its documentation.
• Immersive Pedagogical and Twinning Activities
  This was a course activity in which we were asked to prepare for any electrical engineering topic we like and explain it to students from other colleges. We had taken the topic of Application of Statistics in Digital Signal Processing. You can find the ppt used for the presentation here.