Exploring Quantum Machine Learning Algorithms
Guide: Prof. Anirban Dasgupta | IIT Gandhinagar
- Read papers on quantum machine learning particularly quantum recommendation systems proposed by Kerenidis and Prakash and quantum-inspired classical version of it by Ewin Tang.
- Implemented the classical recommendation system proposed by Ewin Tang.
- Implemented quantum generative adversarial networks (QGAN) in TensorFlow Quantum Framework and trained the QGAN on the MNIST dataset.
Mini-Tweet: A Twitter-like Application Using Socket Programming
Computer Networks Course Project
- Built a mini version of Twitter which supported concurrent client requests from scratch using socket programming.
- Supported features include post and delete tweets, show user feed and support retweeting, show followers and following users, show whether a user is online and offline and showing trending hashtags
- A Mininet emulator was used to test the scalability of the application for concurrent users for different topologies.
Generative Fashion for Indian Clothing
Machine Learning Course Project
Accepted at the Young Researchers' Symposium Track at the ACM India Joint International Conference on Data Science and Management of Data (CoDS-COMAD), 2021
- Curated a dataset containing 12K images pertaining to Indian Context along with descriptions from Myntra and Amazon.
- Used a 2 step GAN model - first one for the generation of the segmentation map and the second one for texture rendering based on textual descriptions.
- Experimented with BERT, TF-IDF, Doc2vec embeddings for textual descriptions.
Pico++ Compiler: A Basic Compiler in C
Compilers Course Project
- Wrote code of a very basic compiler that supports features like if-else, for, while loops including nested versions.
- Defined grammar in Flex to handle various cases and to support these functionalities.
Neural Title Generation for Scientific Papers
Natural Language Processing Course Project
- Created a model involving encoders and attention layers to predict titles given a scientific paper/article.
- Made heuristics to make the predicted title catchy and defined the metric to measure catchiness of the generated title.
- Achieved a BLEU score of 0.54 on the combination of datasets for scientific papers of ACL Anthology and arXiv.
- Showed that the overall pipeline outperforms baselines and is a good direction towards summarization of scholarly documents by the neural approach.
Blind Motion Deblurring through SinGAN Architecture
Guide: Prof. Shanmuganathan Raman | IIT Gandhinagar
Accepted for Oral Presentation at the Deep Internal Learning (DIL) Workshop @ European Conference of Computer Vision (ECCV) 2020
- Proposed architecture for blind motion deblurring through the SinGAN architecture.
- Mainly involved passing image through finer scale GANs in the multi-scale architecture and iteratively upsampling it.
- Achieved a SSIM of 0.52 on the GoPro dataset used for DeblurGAN.
Contributing to GitHub repository vtop
Operating Systems Course Project
- Solved issues posted in the GitHub Issues section of the repository vtop [link].
- Added features like pid of the process, the core in which the process is running, displaying warning on high CPU and memory usage, etc.
- Added new functionalities to vtop to make it more user friendly.
Email Parser for Gmail
HackRush '19 (Hackathon at IIT-Gn)
- Built an email parser to extract data of talks going to take place at IIT-Gn from Gmail on different fields like Computer Science, Maths, etc.
- Used the Gmail API, Stanford Parser and regular expressions in Python to achieve the same.
Survey on Game of Chomps and Angel Game
Discrete Mathematics Survey Project
- Described in detail about two well known combinatorial games - Game of Chomps and the Angel Game.
- Explained about the winning strategies for both the games in various scenarios.
Gesture to Text Converter
Metis - Coding Club
- Built a convolutional neural network classifier using the Kaggle dataset and then used it to predict gestures on new images.
- Took input from the Webcam to predict the gesture and output text using the model trained.
- Obtained 96% accuracy on the data set provided by MNIST.
Interactive Dictionary and Web-Scraping
Computing Course Project
- Wrote a python script - an interactive dictionary that takes the word as input and outputs its meaning. If word entered is not there in data, then it displays the close matches of that word.
- Scraped data of accidents from Wikipedia and python snippets from lecture slides [link] of the course.
- Wrote python scripts to extract data from Airtable and dump data into it.