About Me
I am Mosarrat Rumman, a machine learning researcher
I have expertise in Machine learning and deep learning specially in the field of Computer Vision and Natural Language Processing. I try to stay updated in these fields by reading research papers and trying to implement them by myself. Currently my interest lies on generative models like Diffusion Probabilistic models, Generative Adversarial Networks(GANs) etc. I am also interested in Natural Languae generation using transformer based models. However having particular interest in biological science, I love to work with medical image analysis.
I have worked for almost 4 years in one of the largest banks in Bangladesh, as a Principal officer in the Application Architect and Development team, where I worked as a senior backend developer and database engineer of the Core Banking System. This job gave me the opportunity to build strong expertise in analyzing the huge database of the Core Banking System and writing complicated SQL queries for various reports and performance analyses. However, while pursuing my Master’s alongside this job, I realized my passion and thrill for research works. I was always full of ideas, but rarely had time to work on them while on this job. I love reading research papers and learning new technologies. Therefore, despite having smooth career growth (2 promotions in 3 years), I left the job to pursue my dream of being a researcher.
7+
Awards and scholarships
Research & Publications
I try to share what I have learn. It helps me a lot. Hopefully you will be beneficial.
In this paper, SPECT images of early diagnosed patients and healthy controls are collected from PPMI database. Instead of using a computational heavy CNN, a simple approach is used where only the region of Interest, i.e,. the dopeminergic region is considered and a single perceptron is trained with the are of ROI. The model achieved an accuracy of 94%.The paper was presented in 2018 Joint 7th International Conference on Informatics, Electronics & Vision (ICIEV) and 2018 2nd International Conference on Imaging, Vision & Pattern Recognition (icIVPR) and is publised in IEEE Xplore. The paper has 21 citations till now.
Link to paperParts of speech (POS) tagging is done on Bangla - which is a low resource language. Two POS taggers were built: 1. Using a Hidden Markov Model. 2. Using deep learning model - biLSTM. The performance of both the models were analyzed on mutliple variables. It can be inferred from the results that increasing the size of dataset has greater positive impact on the performace of Bi-LSTM model than on the HMM model.The paper is presented in 2021 IEEE International Women in Engineering (WIE) Conference on Electrical and Computer Engineering (WIECON-ECE).
Link to paperSkills
Python
- Proficient in Tensorflow, Pytorch
- Building REST API using Flask
- Data cleansing and preprocessing using Pandas, OpenRefine
- Data visualization with Matplotlib
Other Programming languages
Database
- Highly skilled in SQL
- Moderately skilled in MongoDB
Language Proficiency
- IELTS 7.5 Breakdown - S: 8, R: 8, L: 7.5, W: 7
Education
MSc in Computer Science
Brac University
CGPA: 3.95 / 4
Projects:- Generating Fashion images using Conditional DC-GAN according to given cloth category
- Extractive summarizer using BERT
BSc in Computer Science & Engineering
Brac University
CGPA: 3.51 / 4
Projects:- IoT based weather station (Arduino, sensors and google firebase)
- Object detection for blind people (Raspberry pi, sensors, Google Vision API)
- Hospital Management System Web Application (PHP, MySQL, HTML, CSS, Bootstrap)