Bangladesh University of Engineering and Technology

CGPA: 3.75 out of 4.00 (obtained Marks: Above 85%), thesis ongoing.

Thesis on Geo-spatial understanding of LLM

Bangladesh University of Engineering and Technology

CGPA: 3.77 / 4.00 (3.89 in senior year)

Thesis: A MULTI-MODAL DEEP LEARNING BASED APPROACH FOR HOUSE PRICE PREDICTION

1. Md Hasebul Hasan, Md Abid Jahan, Mohammed Eunus Ali, Yuan-Fang Li and Timos Sellis. A Multi-Modal Deep Learning Based Approach for House Price Prediction. Data Mining and Knowledge Discovery (submitted, April 2024) manuscript pdf


2. Md Hasebul Hasan, M. S. Islam Khan, M. S. Haque, A. Y. Ahmed and M. A. Adnan, GuardFL: Detection and Removal of Malicious Clients Through Majority Consensus, 2024 IEEE 48th Annual Computers, Software, and Applications Conference (COMPSAC), Osaka, Japan, 2024, pp. 1112-1117, doi: 10.1109/COMPSAC61105.2024.00150. keywords: {Training;Measurement;Accuracy;Federated learning;Benchmark testing;Software;Robustness;Federated learning;backdoor attack;image classification}, paper link


3. Mahir Labib Dihan, Md Tanvir Hassan, Md Tanvir Parvez, Md Hasebul Hasan, Md Almash Alam, Muhammad Aamir Cheema, Mohammed Eunus Ali, Md Rizwan Parvez, Mapeval: A map-based evaluation of geo-spatial reasoning in foundation models,Submitted to The Thirteenth International Conference on Machine Learning, Accepted, 2025, url

Geo-spatial understanding of LLM

• To investigate the geo-spatial understanding of Large Language Models (LLMs) and develop methods for improving their ability to comprehend spatial information in natural language.
• To evaluate the effectiveness of combining multi-agent with LLMs in improving the performance of geospatial queries.

LLM Model with RAG for fetching exact information from structure and unstructure data

• Develop and evaluate a Large Language Model (LLM) that utilizes Retrieval- Augmented Generation (RAG) technology for extracting precise and relevant information from both structured and unstructured data sources.

Enhance Partial Dominant Pruning for AD HOC Wireless Networks

• Feature: AD HOC Networks, Dominant Pruning
• This project explores Enhanced Partial Dominant Pruning (EPDP), an enhancement proposed to address issues with Partial Dominant Pruning. We evaluate its performance and analyze its potential for further improvement.

Attacking and Defending ML Application on Public Cloud

• Feature: Cloud security, AWS EC2, adversarial attack, CNN
• This project proposes a framework to protect neural network classifiers from adversarial examples using detector and reformer networks, randomness, and diversity. It effectively addresses challenges in defending against advanced attacks without sacrificing the false positive rate on normal examples.

Classifying DNA Sequence

• Feature: Bioinformatics, Machine Learning
• This project uses machine learning and deep neural networks to classify short E. Coli DNA sequences by employing various classifiers. The dataset consists of 106 DNA sequences, each with 57 nucleotides, sourced from the UCI machine learning repository.

Cluster Based Influence Detection in Worldwide Trade Data

• Feature: Network Science, NetworkX,
• In this project, we use centrality measures and DeepWalk for clustering to identify the most influential country in separate communities of a network. We cluster communities based on similar economic trends using international trade data and calculate the influence measure for each country in each cluster using embeddings. We construct a graph using Gephi software and utilize NetworkX package to rank the countries from each cluster and identify the most influential countries.

Food Image Classification

• Feature: Deep learning, Transfer learning, CNN, Resnet50
• I along with my partner build a CNN model to classify the food images into 10 classes. We train our model using 10k food images. We also fine-tune some pre-trained models (Resnet50, Efficientnet B7, Inception V3) and show their performance statistics.

Convolution Neural Network(CNN) From SCRATCH

• Feature: Deep learning, CNN
• Implemented a convolutional neural network from scratch with python for an image classification task with support for variable number of layers, dimensions as well as batching.

HIDDEN MARKOV MODELS

• Feature:Python, Probabilty
• Hidden Markov Model (HMM) implemented with the Viterbi algorithm.
• The parameters of the model was estimated with Baulm-Welch Algorithm

Security Project

• AES standard implementation: An implementation of AES by python from scratch. This tool can encrypt as well as decrypt texts and files of any kind.
• Cross-Site Scripting (XSS) Attack: Used a web application named Elgg provided in a pre-built Ubuntu VM image and exploited XSS vulnerability to launch an XSS attack on the modified Elgg.
• Dos-Attack-ON-Dns-Server-using-ip-Spoofing: Here I implement Dos attack on Dns server (using ip spoofing) . For this attack I need three machines. One machine works as Dns server and others are the attacker machine and the normal user machine .Then From the attacker machine I execute a c code which performs a query flood attack to our DNS server . This dos attack affects our DNS server.DNS server stops to receive requests .So normal user can’t visit there requested page because the ip address of their requested page can't be translated.

Amader Internet

• Features: Software Development, Vue.js, Node.js, Mongodb
• It's basically a website built for internet users, ISP, and NTTN in Bangladesh motivated by the info Sarkar phase-3 project. This website was built along with two of my partners using Vue-js as front-end and Node-js as back-end.

Classical AI Projects: Ghost Hunter

• AI-game-LinesOfAction: Used minimax algorithm with alpha-beta pruning with suitable heuristics to create and agent which is able to play the board game. Can also be played in multiplayer mode.
• Ghost Hunting: Modeled Ghost Hunting problem using HMM and Particle Filtering.
• Latin Square CSP Solver: Used several CSP solving algorithms with heuristics to solve latin square.
• Optimal Exam Scheduling with local search: Exam scheduling problem was solved by local search with appropriate heuristics.
• Solving n-puzzle by A-star search: Solved n-puzzle problem with A-star search by using two heuristics (the misplaced tiles an the Manhattan distance).

Online Service Platform

• Feature: Java, Postgresql, relational database, Java Fx
• This desktop application was built using JavaFX for UI and Postgresql for the database. Here we mimic ShebaXYZ.

Hospital Management System

• Feature: Java, Networking, Java Fx, Multi-Threading
• It is a management system that provides many functionalities for the patients, doctors, and authorities. Based on multi-threaded and networking, this software provides access for many patients and doctors at a time.