Semi-Supervised Learning using Graph Random Neural Networks
Semi-Supervised Learning using Graph Random Neural Networks

Implemented a GNN architecture much less prone to over-smoothing, much more robust to graph attacks and much better at generalization in semi-supervised settings where availability of labelled nodes is scarce.

Graph Neural NetworksSemi-Supervised Learning
Wind Energy Forecasting
Wind Energy Forecasting

Performed an inter-state comparison for the best place to set up a wind farm by performing time-series forecasting using traiditional statistical methods and deep learning algorithms.

Time Series ForecastingARIMAML
Content-Aware Image Resizing Using Seam Carving
Content-Aware Image Resizing Using Seam Carving

Implemented the classic seam carving algorithm using a minimum energy path approach using dynamic programming. Applied both Sobel and Laplace filters for generating the energy map over which the algorithm was run.

Image ProcessingDynamic Programming
Handwritten Digit Recognition using Capsule Networks
Handwritten Digit Recognition using Capsule Networks

Developed a solution for indigenous handwritten digit recognition by implementing and augmenting the novel Capsule Networks architecture.

Computer VisionCapsule Networks
Image-to-Image Translation using GANs
Image-to-Image Translation using GANs

Developed a GAN-based solution(DualGAN, CycleGAN) for performing unsupervised, unpaired image-to-image translation. Improved qualitative results and training stability by augmenting the model architectures.

Computer VisionGAN