Sudipta Progga Islam

Education

Rajshahi University of Engineering & Technology (RUET), Rajshahi, Bangladesh
Bachelor of Science in Computer Science and Engineering
2019 - 2024 | CGPA: 3.84 out of 4.00 | Position: 4th out of 179

Relevant Coursework: Operating Systems, Applied Statistics & Queuing Theory, Artificial Intelligence, Neural Networks & Fuzzy Systems, Data Mining, Digital Image Processing, Database Systems, Parallel and Distributed Processing, Digital Signal Processing, Network Security, Software Engineering

Standardized Test Scores

IELTS: Overall 7.0 (Listening: 7.5, Reading: 7.5, Writing: 6.5 & Speaking: 7.0)

Publications

Optic Disc and Cup Segmentation via Enhanced U-Net with Residual and Attention Mechanisms

ICEEICT 2024 | IEEE Xplore | DOI: 10.1109/ICEEICT62016.2024.10534436

• Evaluated various pretrained models as U-Net backbones, validated across Drishti-GS, REFUGE, and RIM-ONE-R3 datasets, and developed an enhanced U-Net with residual and attention mechanisms.
• Award Nomination: Nominated for Best Poster Award at ICEEICT 2024.

Advancing Ophthalmology through Transfer Learning and Channel-wise Attention for Retinal Disease Classification

ICEEICT 2024 | IEEE Xplore | DOI: 10.1109/ICEEICT62016.2024.10534342

• Developed a hybrid model merging EfficientNetB0 and InceptionV3 with channel-wise attention, improving discriminative ability by dynamically adjusting attention across channels, outperforming state-of-the-art models.

BanglaOngko: A New Dataset for Accurate Bengali Mathematical Expression Detection Utilizing YOLOv8 Architecture

BIM 2023 | Taylor and Francis | Project

• Created and annotated the BanglaOngko dataset with Roboflow, developed an efficient algorithm integrating statistical concepts to accurately localize handwritten Bengali mathematical expressions, addressing YOLOv8’s unsorted bounding box challenges.

A Deep CNN-Based Approach for Revolutionizing Bengali Handwritten Numeral Recognition

BIM 2023 | Springer | DOI: 10.1007/978-981-99-8937-914

• A task-oriented deep convolutional architecture for recognizing Bengali handwritten numerals, with a focus on achieving high accuracy while using a relatively small number of parameters.

Undergraduate Thesis

RAD-U-Net: Residual Attention-based Dual-path U-Net for Segmentation of Multiple Myeloma Cells in Microscopic Images for Clinical Diagnosis [Project]

Python, Tensorflow, Keras, CNN, Attention mechanism, Residual connection, U-Net

• Designed and implemented RAD-U-Net, a dual-path U-Net model enhanced with attention and residual connections for precise segmentation of myeloma cells. Combined low-level and high-level features for improved accuracy and computational efficiency, enabling practical use in clinical settings.

Projects

Optimizing Feature Representation of Deep Neural Networks for Enhanced Deepfake Detection [Project]

VGG16, Channel Attention, ResNet50, Deepfake

• Focused on detecting deepfake images using the 140k Real and Fake Faces dataset. Employed VGG16 as the primary feature extractor, then introduced a channel attention mechanism to prioritize relevant feature vectors and finally achieved a high classification accuracy of 99.80%. Further, conducted an ablation study with ResNet50.

Fake News Detection System [Project]

NLP, Tensorflow, Keras, GloVe Embeddings, LSTM

• Built a system to classify news as fake or real using GloVe embeddings and an LSTM-based neural network in TensorFlow. Performed data preprocessing and analysis, and evaluated the model with accuracy, confusion matrix, and ROC curve.

Cardiotocogram Data Analysis for Fetal Health Classification Using Machine Learning [Project]

Random Forest, K-Nearest Neighbors, Gradient Boosting, SMOTE

• Aimed to classify fetal health status from Cardiotocogram (CTG) data using machine learning models. Utilized a dataset of 2,126 records with 21 features. Applied models like Random Forest, K-Nearest Neighbors, and Gradient Boosting, achieving the highest accuracy of 98.47%. Employed feature selection, data standardization, and SMOTE to enhance performance.

Maternal and Child Health Care [Project]

HTML, CSS, PHP, MySQL, Android Studio, Java, XML, Firebase Database

• Developed a responsive website for Maternal and Child Health Care featuring due date calculations, immunization schedules, personalized notifications, and a query posting feature. Created a mobile app using Android Studio and Firebase with the same features.

Library Management App [Project]

Android Studio, Java, XML, Firebase Database

• Developed a mobile application to streamline library operations, including catalog management, user borrowing, and return tracking.

Technical Skills and Interests

• Research Areas: Computer Vision, Medical Imaging, Object Detection, Unsupervised Learning, Data Science, Natural Language Processing, Large Language Models, Vision Language Models
• Programming: Python, C, C++, Java, JavaScript, PHP
• Frameworks: TensorFlow, Scikit-Learn, Keras, OpenCV, PyTorch, Bootstrap
• Web & Databases: HTML, CSS, PHP, MySQL
• Technologies: Flask, Android Studio, LaTeX, Git

References

Dr. Md. Nazrul Islam Mondal (Undergrad Thesis Supervisor)
Professor
Department of Computer Science & Engineering
Rajshahi University of Engineering & Technology
Mobile: +880-17206622788
Email: mondal@cse.ruet.ac.bd

Nahin Ul Sadad (Project Supervisor)
Assistant Professor
Department of Computer Science & Engineering
Rajshahi University of Engineering & Technology
Mobile: +880-1778059702
Email: nahin@cse.ruet.ac.bd

Contacts

Phone: +880 1842-090214
Email: proggasudipta0@gmail.com
LinkedIn: Sudipta Progga Islam
Address: 404, 4C, Rajuk Uttara Apartment Project, Sector-18, Uttara, Dhaka-1230, Bangladesh