Aditya Panda

I am currently a final year doctoral student at Indian Statistical Institute. During the last few years, I mainly worked on data efficient deep learning approaches. Specifically I explored the different aspects of the problem Compositional Zero-Shot Learning (CZSL). Compositional Zero-Shot Learning, focusses on recognizing new state-object compositions from limited training data. We worked on approaches using graph convolutional networks, and transformer networks to tackle challenges in the closed-world, open-world, and partially supervised CZSL problems. Before my PhD, I completed my M. Tech degree in Computer Science from ISI, Kolkata. Prior to that I pursued my undergraduate studies in Electronics and Tele-Communication Engineering from Jadavpur University, Kolkata.

Email  /  Google Scholar  /  CV  /  Semantic Scholar /  Orcid Id /  IEEE Author Profile

Publication List

In this work, we proposed a Knowledge-Guided Transformer Network to model the partial associations between visual features of state and object, to better tackle the intra-class variability in unseen state-objcet compositional images.

In this work, we proposed Graph Convolutional Network based approach in attempt to solve the open-world CZSL problem..

In this work, we attempt to better tackle the intra-class variability in unseen state-objcet compositional images by using a Sequential Learning approach.

In this work, we attempt to disentangle the visual features of state and object in a state-objcet compositional images.

Manuscripts Under Review