Uploaded folders contains all the projects that I did during Spring 2021. The following paragraph contains a brief information about what each of the folder contains. For more details about each project, please browse each folder.
My primary project SG-Net-Master is an implementation of this paper https://arxiv.org/abs/1908.05147 which was published in 2019. The code was initially forked from the github of the author and then modified accoridng to my goal. The descriptions will be found inside the folder.
The assignments contains reviews of a paper titled "Visualizing the Loss Landscape of Neural Nets" and a short summary of 3 papers which participated in the SQuAD 2.0 challenge. The Jupyter notebook visualizes how the loss surface diverges to local minimum. It also contains some slides that I presented.
This project was done in order to learn how to use pytorch transformers library. In this project I used the transformer library as an intact module, and also used library of encoder and decoder to form a transformer to do the same thing.
BERT Transformer user the pretrained library from the Huggingface transformers library. The goal of this project is to do sentiment analysis using Pre-Trained BERT.
This code is forked from stanford NLP where they implemented the Transformer without using any build in library. They build the model and tested it using an available dataset from torchtext dataset.
NeMo(Meural Moduler) tutorials are imported from their github repository. This project was done solely to learn how to use the NeMo Library. Here I learned about 2 modules. One is Speech to Text and another is Qustion Answering using SQuAD 2.0 dataset.
This project was done as a part of complesion of the Deep Learning course by Udacity. I completed three modules along with the given assignments. All the coding and and assignments could be found in the respective folder, naming, 'Intro to Pytorch', 'CNN' and 'RNN'.
This was a side project colaborating with another student. The goal of this project was to integrate Zoom SDK with an web application and the web application will be automatically able to initiate a call or join a meeting given a schedule. Implemented this task using a standalone python application and also using a Node web application.