Projects
Learning Robust feature representation
[August 2020 - Present, ICLR CSS track]
Proposed and got acceptance for RFP(Research for plot) draft for CSS track of ICLR 2022. The project is underway, find the details here.
[Oct 2019] [GitHub here]
Implemented and trained the model described in the paper Brain Tumor Segmentation with Deep Neural Networks(Havaei et al,2017). The project was done solely, implemented on Keras in Python.
For this 2D slice-based segmentation approach, designed a slice selection scheme discarding blank and just non-tumor class present slices, for efficient training on limited compute. Owing to the same, implemented weighted cross-entropy loss for tackling class imbalance.
This project was done as part of a research internship under Prof. Aditya Abhyankar, Dean of DoT, Pune University.
Artist Identification
[Sep 2018] [GitHub here]
Implemented artist identifier as a 12 class classifier according to this project report. Extending this implementation, proposed a siamese network for flexibility of adding classes. Further proposed a siamese network with style loss to capture a domain-specific objective. Implemented on Keras in Python.
This project was done in partial fulfillment of the course Neural Networks and Fuzzy Logic in BITS Pilani; solely worked on the further extensions.
SIGNATURE VERIFICATION USING CNN
[January 2018] [GitHub here]
Worked on applying deep learning for signature verification. Implemented verification as multi-class classification using novel architecture and ResNet50. Trained a NasMobileNet with Siamese training and triplet loss.
This project was done in partial fulfillment of the course study-oriented project in BITS Pilani.
FAKE NEWS DETECTION
[November 2017]
Implemented LSTM architecture to detect biases in news articles from Indian news vendors and US presidential election dataset from Kaggle.
This project was done in partial fulfillment of the course information Retrieval in BITS Pilani.
Logic Programming
[November - December 2016]
Implemented 2 tasks using the logic programming paradigm in SWI-Prolog. The first task was to encode rules for performing basic algebraic manipulations by operating on symbols. The second was to encode the academic regulations of BITS Pilani in the form of rules which could then answer complex queries.
This project was completed for the partial fulfillment of the course Logic in Computer Science in BITS Pilani.