projects | Karish Grover

nlproc

This project considers the fake news detection problem under a more realistic scenario on social media. Given just a source short-text tweet, we aim to predict whether the given tweet is fake or not.
We further aim to generate an explanation for this prediction. This paper presents Attention-enhanced Multi-channel Recurrent Convolutional Network (AMRCN), for explainable fake news detection.

Verifying the integrity of a particular piece of news is not easy for any end-user. In this work, our end goal is to design an efficient model to examine the reliability of a news piece.
This paper addresses the problem of Fake News classification, given only the news piece content and the author name.
We present, Hierarchical Convolutional-Attention Network (HCAN) composed of attention-enhanced word-level and sentence-level encoders and a CNN to capture the sequential correlation.

A deep learning based model for the task of measuring cross-lingual and multi-lingual news article similarity.
Two parallel pipelines:- graph-based (Multilingual abstract meaning representation for knowledge graph-level news matching and text-based (Multihead attention over multilingual BERT for text-level news matching).

This work proposes a solution for expression- and orientation-aware facial unmasking. To the best of our knowledge, this is the first attempt to generate unmasked faces while preserving the person’s emotions and facial orientation.
We propose a novel GAN-based approach, enhanced with Fourier convolutions to generate a high-resolution unmasked image.
Further, we propose a dilated-convolutional refinement module that enhances the decoder output quality, to remove noise from the generated image.

An interactive tool that performs fine-grained analysis and visualize diffusion patterns on real-time Twitter data.
Classify content into categories like authentic, fake, satire, imposter, manipulated, hatred, and misleading content.
We train our tool on the state of the art deep learning models and further test them using real-time Twitter data and offer a number of visualizations like diffusion of content through Twitter, polarization in the network, etc.

Developed an application that manages the data of various aspects of a law firm, like details of lawyers, their ongoing cases, employee earnings, partner earnings, etc.
Suitor supports queries like finding the best suited lawyer on the basis of track record, time taken to solve cases, fee etc, and helps young associates to find their career forte on the basis of theirtrack record while working on cases in various legal fields.
We construct the schema and entity relationship diagrams and create sample dataset tables with entries of over 10000 employees.

Created a revamped clone of the mobile game, Color Switch.
Implemented a game engine from scratch in JavaFX CSS and FXML with optimized game-play for smooth rendering.
Utilized design patterns, UML diagrams and OOPs. Created new obstacles with custom animations, that increase in difficulty according to the efficiency of the Player.