Hello and Welcome! I am ,Md Adith Mollah (aka ,'Adith') from Dhaka, Bangladesh, graduated ( in 2024) from the department of Computer Science and Engineering (CSE) at Shahjalal University of Science and Technology (SUST) Sylhet, Bangladesh by obtaining BSc. degree. I am Extremely passionate about conducting research related to Trustworthy Machine Learning. I love to solve and explain coding problems, explore new techs and research topics, read (fiction, non-fiction) books, travel and click unique snapshots of nature.

My recent research work depicts about inspecting different versions of a particular black-box (Deep Learning) models predicted pixel regions (of BrainTumor images) comprehensive analysis (trust worthiness). In addition, I have several other published (conference, arXiv) research works related to Natural Language Processing and Computer Vision. Moreover, my undergraduate thesis is about proposing a system that leverages Deep Learning techniques to generalize deep fake image detection.

From march 2024, I joined as a Graduate Engineer in a research and development company known as 'Kona Software Lab Limited' . Always open to collaborate with any type of research work, startup idea and educational content creation.

Email: adibhasan35 (at) gmail (dot) com

Photo taken at Cox's Bazar, Bangladesh, the longest natural sea beach in the world

CV /  LinkedIn  /  Google Scholar  /  ResearchGate  /  GitHub  /  YouTube  /  X

News and Updates

Research Experience

My current research interest lies broadly in Trustworthy Machine Learning . More specifically, I am interested in utilizing novel and existing Explainable Artificial Intelligent or XAI techniques to inspect black-box models transparency as well as constructing trustworthy black-box models specially in healthcare, agriculture, privacy and security. I aim master 'model interpretability and explainability', 'fairness', 'adversarial machine learning', 'differential privacy' and 'causality' which can be which are part of trustworthy machine learning family.

The following sections depict my involvement in research works related to Trustworthy Machine Learning, Deep fake image detection (undergrad thesis) and several (Natural Language Processing and Computer Vision) works,

Generalized Fake Image Detection Model using Deep Learning Techniques

Supervisor(s):

Leveraging LIME to Inspect Trust Worthiness of Custom Pretrained Xception Transfer Learning and Fine Tuning Variants on Brain Tumor Image Classification

Provided a compehensive analysis to figure out which approach of proposed Xception models is the most trust worthy by inspecting pixel regions that contributed to predict particular brain tumor image.

Exploration of Transformer Ensemble and Auto Regressive Approaches to Enhance Performance of Clickbait Title Detection

In this paper, we tried to provide model performance comparison by training on both balanced and imbalanced version of clickbait text data. Based on the analysis, we provided future directions regarding the improvement of clickbait title detection.

SUST_Black Box at BLP-2023 Task 1: Detecting Communal Violence in Texts: An Exploration of MLM and Weighted Ensemble Techniques

we leveraged masked language modeling and weighted ensemble techniques for detecting communal violence in Bangla texts

Supervisor(s):

Performance Enhancement Leveraging Mask-RCNN on Bengali Document Layout Analysis

We utilized step by step hyper-parameter tuning to improve performance on mask-RCNN model which was trained on BaDLAD dataset

Projects

This section includes competition and several notable undergraduate course projects.

AI-Agent-Debug

This project demonstrates an AI agent that interprets user queries to identify the appropriate tools via a language model (LLM) and combines the results from these tools(Foreign Exchange Rates, Calculator, Weather, Knowledge-Base) with the LLM to generate accurate, reliable, and non-hallucinated responses.

AgriDoctor

We built a Java web application focused on enhancing agricultural domain by integrating AI models. This web application is the composition of 3 core features known as Crop Recommendation, Fertilizer Recommendation, Plant Disease Detection

Deep Fake Image Detector

A mobile application, able to detect deep fake images providing confidence rate. More specifically, the application is able to detect between style GAN1, original and style GAN2, original images. Xception, CNN architechture is utilized for making the classifier.

Tomato Blight Detector

CNN model detecting Tomato Early Blight and Late Blight disease by capturing images from mobile

Subtitle Hack

Main focus of this project is to familiarize the users with English Vocabulary as much as possible efficiently, saving the search time on search engines.

Count If You Can

An Interactive Java desktop application which helps the user to test and improve memorizing skills

Education

Work Experience

Honors and Awards

Have been involved with some of the honors and awards which are mentioned below. Extremely thankful to my team-mates for their relentless contribution.

Leadership

The following sections denote my involvement in leadership programs and activities

Content Creation

The following sections denote my involvement in actitivies that I love to do during free time with the intention of giving back to community.

Special thanks to my friend Asif Haider for the code template that helped to build this portfolio website