Projects
This page is NOT being updated and only archives older projects.
Projects up to 2022
- Investigating the Application of VLAD to NLP (case study on Irony/Sarcasm Detection in Twitter)
- Venue: Dept. of Computer Science, Colorado State University
- Description: Investigating the application and effectiveness of vector of locally aggregated descriptors on top of Transformer-based language representation layers; Stuyding Irony/Sarcasm detection as a use case.
- Social Media Bot Identification using Deepfake Text Detection
- Venue: Dept. of Computer Science, Colorado State University
- Description: Applying Transformer-based models to detect bot-generated text on a deepfake dataset.
- Claim/Counterclaim detection in Social Media
- Venue: Dept. of Computer Science, Colorado State University
- Description: Designing a pipeline to detect claim and counterclaims in social media incorporating Natural Language Inference, Stance Detection, and Claim Detection; In Twitter and Youtube; Working on publishing an annotated COVID-19-related claim/counterclaim dataset from YouTube comments.
- Automatic Access Control Policy Extraction from Natural Language Requirement Documents
- Venue: Dept. of Computer Science, Colorado State University
- Description: Currently writing a proposal for this project. We aim to use a transformer-based semantic role labeling model to extract policy information from natural language text and translate them to Next Generation Access Control (NGAC) graph representations.
- Machine Translation for Similar Low-Resource Language Pairs with Loan Words
- Venue: Dept. of Computer Science, Colorado State University
- Description: Currently studying the potential benefit of using load words in similar languages to incorporate that knowledge into a machine translation framework.
B. Sc. Project
- Implementation of a Convolutional and a Pooling Layer of a CNN on FPGA
- Venue: Dept. of Computer Engineering and Information Technology, Amirkabir University of Technology
- Description: In this research, I implemented the convolutional and the max pooling functions of CNN’s using Xilinx Vivado High-Level Synthesis and ran this project on a ZYBO SoC board. This design acheived up to 30 times faster throughput relative to the software code on a CPU.
- Notes: The English abstract is available here. My dissertation and the presentation slides are available in Persian.
Other Projects
- Named Entity Recognition for Persian Language
- Venue: Dept. of Computer Science, University of Nevada, Las Vegas
- Description: Implemented using biLSTM-CRF architectures for Persian language which is a low resource langugae.
- Activity Recognition with Wearable Sensor Dataset in Spark Platform
- Venue: Dept. of Computer Science, University of Nevada, Las Vegas
- Description: Implemented scalable classification solutions (Multilayer Perceptron Network, LogisticRegression, and Decision Tree) using Spark libraries for motion data of elderly peoplein a room.
- Project report manuscript is available here
- Multithreaded Persian Short-Text Classification
- Venue: CommentMiner
- Description: Implemented using two different algorithms: MaxEnt and Naive Bayes, on the datasets mentioned above with the accuracy of 70%, (Java).
- Key-phrase Extraction on Persian Documents
- Venue: Dept. of Computer Engineering, Sharif University of Technology
- Description: Literature review was carried out on KP-Miner, TF-IDF, WINGNUS, Single Rank, KEA, Topic Rank and Salience Rank methods. We used Inspec and 500N datasets to test these methods and then worked on Salience Rank model on a Persian news dataset.
- Telegram Chat-bot for Automatic Question Answering
- Venue: CommentMiner
- Description: Implemented as a first and simple model using indexing and similarity checking on: Who Is Hosting Who FAQ and on Question-Answer Jokes, (C#).
- Template Matching
- Venue: Dept. of Computer Engineering and Information Technology, Amirkabir University of Technology
- Description: Implemented using both OpenMP and CUDA to parallelize the algorithm and to enhance the performance, (C, C++).
- Hand-written Digit Image Generation
- Venue: Amirkabir University of Technology
- Description: Implemented using a simple Generative Adversarial Network on MNIST dataset, (Python Keras package).
- Hand-written Digit Image Classification
- Venue: CommentMiner
- Description: Implemented using a Convolutional Neural Network on MNIST dataset, (Python Tensorflow package).
- Sequential MNIST Classification
- Venue: Deep Learning Summer School at University of Tehran
- Description: Implemented using a Recurrent Neural Network on sequential images (28 sequences) of MNIST dataset, (Python Keras package).
- Persian Short-Text Sentiment Analysis
- Venue: CommentMiner
- Description: Implemented as a first model using two different algorithms: MaxEnt and Naive Bayes classification to positive, negative and neutral classes, on Instagram comments, (Java).
- A Genetic Soloution to Travelling Salesman Problem
- Venue: Amirkabir University of Technology
- Description: Implemented as the course project to find the best route among 24 random cities, (Matlab).
More projects grouped by their PL:
- Python:
- Multilayer Perceptron classifier with back propagation using pure python to classify 3 types of seeds
- Java:
- Re-implemenation of Robokill Game
Robokill gameplay image taken from SilverGames - Internet Messanger (with GUI design)
- Internet Browser with the support for PHP and CSS using JxBrowser library
- Notes Manager
- Tank Game (two tanks attacking each other)
- Sudoku Solver
- Box Stacking Problem Solver
- Thief and Cops Game (console)
- Insertion Sort, Heap Sort, Quick Sort, Merge Sort
- N-equation Solver
- Re-implemenation of Robokill Game
- C/C++/C#:
- OpenMp Merge Sort
- CUDA Matrix Multiplication
- CUDA Prefix Sum
- 2048 Game (console)
- Game of Life (console)
- CommentMiner Main System (in collaboration with other teammates)
- Web:
- Re-implementation of IranFilms that is a website for introducing and reviewing movies
- PHP:
- Telegram Bot to provide CommentMiner services response to a private message sent by the user