Notable Projects
Denoising images with Convolutional Autoencoder.
In this project, I utilized Convolutional Autoencoders to denoise images by learning efficient low-dimensional representations. The model was trained to reconstruct clean images from noisy inputs, leveraging convolutional layers to capture spatial features. Key steps included dataset preparation, network architecture design, and performance evaluation using metrics such as PSNR and SSIM to ensure high-quality image restoration.
Project Link: Github Repository
Computational-Statistical Tradeoffs in learning Graphical models
In this project, I investigated the computational-statistical tradeoffs involved in learning graphical models, focusing on balancing algorithmic efficiency with statistical accuracy. The work involved developing methods that optimize learning performance under resource constraints while ensuring robust representation of underlying data dependencies. Key contributions included theoretical analyses and the implementation of scalable algorithms for structured learning tasks.
Project Report: Pdf
Structure Learning in Gaussian Graphical Models
In this project, I explored various popular algorithms for Gaussian graphical model structure learning. The focus was on evaluating methodologies used to identify conditional independence relationships and estimate sparse precision matrices, particularly in high-dimensional data settings. The study aimed to provide insights into the strengths and limitations of these approaches.
Project Report: Pdf
Tensor SVD: Statistical and Computational Limits
Singular value decomposition (SVD) and principal component analysis has been an important tool in multivariate and high dimensional data analysis and have been thoroughly studied in the case of matrices, however they only capture first order interactions and ignore higher order ones. In this project we explore the statistical and computational limits for the Tensor SVD problem.
Project Report: Pdf
An Information-Theoretic Approach towards Understanding the Utility-privacy Tradeoffs in Databases.
In this project, I explored the utility-privacy tradeoff in databases through an information-theoretic lens. The study focused on (theortically) quantifying how privacy-preserving mechanisms impact the utility of shared data, aiming to develop frameworks that balance these conflicting objectives. Key aspects included analyzing privacy constraints, designing mechanisms with provable guarantees, and applying theoretical insights to real-world database scenarios.
Project Report: Pdf
Implementing Recurrent Neural Network (RNN) and Long Short Term Memory (LSTM) for Electric Vehicle charging demand prediction.
In this project, I implemented Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) models to predict Electric Vehicle (EV) charging demand. The focus was on capturing temporal dependencies in the data to accurately forecast charging requirements. Key steps included preprocessing time-series data, designing and training deep learning models, and evaluating their performance to optimize predictions for real-world applications.
Project Link: Github Repository
Finding optimal policy using Policy Iteration for Electric Vehicle charging demand prediction.
Increasing growing of electricity demand and environmental issues bring huge incentives to electric vehicles (EVs) market. EVs will improve the functionalities of present power system. On the other hand, unscheduled high penetration of EVs may have detrimental effects on power system performance. This project studies the electric EV charging scheduling problem under a charging station scenario, aiming to offer an optimal policy to optimize the battery configuration based on load prediction.
Project Link: Github Repository
Support Vector Machine classifier
The focus of this project is to utilize SVM’s ability to maximize the margin between data classes, ensuring robust generalization and accurate predictions. Key steps included feature selection, kernel optimization, and hyperparameter tuning to achieve optimal performance.
Project Link: Github Repository
Implementing the tabular Q-learning algorithm for three standard control problem mountain car, cart pole, and acrobat.
In this project, I implemented the tabular Q-learning algorithm to solve three standard control problems: Mountain Car, Cart Pole, and Acrobot. The work involved designing state-action value tables, updating policies using the Q-learning update rule, and evaluating the algorithm’s performance in achieving optimal control across these environments. Key challenges included balancing exploration and exploitation and handling the complexity of continuous state spaces.
Project Link: Github Repository
Balancing between Electric Vehicle Charging Station Income and Users Cost using Reinforcement Learning
Increasing growing of electricity demand and environmental issues bring huge incentives to electric vehicles (EVs) market. EVs will improve the functionalities of present power system. On the other hand, unscheduled high penetration of EVs may have detrimental effects on power system performance. This project studies the electric EV charging scheduling problem under a charging station scenario, aiming to offer an optimal policy to optimize the battery configuration based on load prediction.
Project Report: Pdf