Publications

2019

Syed Fahim Ahmed, Fairuz Shezuti Rahman, Tasmia Tabassum, Md. Tariqul Islam Bhuiyan, "3D U-Net: Fully Convolutional Neural Network for Automatic Brain Tumor Segmentation", 2019 22nd International Conference on Computer and Information Technology (ICCIT 2019), Dhaka, Bangladesh, DOI: 10.1109/ICCIT48885.2019.9038237

In Progress

Syed Fahim Ahmed, Jixian Li, Mingzhe Li, Bei Wang, Visualizing Activation Spaces of Morse Complex Generation (In Progress)

Projects

Professional Projects [2020-Present]

• VoIP Phone System (May 2021 - April 2022):

  • The IP phone system allows users to make phone calls through an internet connection instead of a regular landline.
  • The system provides advanced functions such as priority call, attended transfer, absence transfer, secretary call, hotline, etc.
  • Tools: Asterisk, Kamailio, Drachtio, Flexisip, PJSUA, Node.js.



• Ohori Koen Kindergarten Management System (January 2021 - April 2021):

  • Developed a web application to manage daily activities in the kindergarten, assisting teachers and parents with the health tracking of children.
  • Provided user-friendly interfaces for checking children's previous and current health statuses.
  • Tools: Laravel, React, PostgreSQL.



• Kyushu Corp. Attendance System (July 2020 - December 2020):

  • Built a system with IoT-enabled real-time face recognition (FaceNet) and IC card reading, integrated with a web server.
  • The system was modified to reduce COVID-19 risks in the company.
  • Tools: CodeIgniter, Bootstrap, JavaScript, PostgreSQL, Python, OpenCV, TensorFlow, Vega3000, Vega5000, Raspberry Pi.



• FAQ System (May 2020 - June 2020):

  • Developed a web application for industrial websites to enhance customer service by providing dynamic search options and statistical data to service administrators.
  • The application helps users find answers to their questions easily and efficiently.
  • Tools: Laravel, Bootstrap, JavaScript, Ajax, jQuery, Chart.js, MySQL.



• Nursing Assistant System (Speaker Identification, Database) (February 2020 - March 2020):

  • Automated the working environment for hospital nurses by developing a speaker identification model using SincNet and designing a relational database.
  • Improved the efficiency of communication and data management within the hospital environment.
  • Tools: PyTorch, MySQL.

Selected Academic Projects

• Benchmark of Deep Learning Algorithms for Left Atrium Wall Segmentation (Fall 2022):

  • Explored multiple deep learning architectures for the segmentation of the left atrium to aid in the treatment of atrial fibrillation.
  • Utilized a range of state-of-the-art models including UNet, UNet++, MANet, and DeepLabV3 to improve the accuracy of medical imaging segmentation.
  • Employed various data pre-processing techniques and advanced metrics like Dice Score and IoU Score for performance evaluation.
  • Project Code: [Code]



• Exploring Seismic Over Time: An Interactive Visualization (Spring 2023):

  • Designed and developed an interactive web-based visualization tool to analyze seismic data trends using JavaScript, Leaflet, and D3.js, enhancing understanding of earthquake patterns globally.
  • Conducted in-depth analysis of seismic activity data, integrating real-time data with historical earthquake information to identify regions at high risk and evaluate disaster impact.
  • Demonstrated project's utility through case studies of major seismic events, providing insights for disaster management and infrastructure planning.
  • Demo Website: [Demo]



• HepaticVision: Deep Learning for Liver and Tumor Segmentation in CT Scans (Fall 2023):

  • Developed HepaticVision using U-Net and Attention U-Net architectures to segment liver and tumors from CT scans, achieving an F1 score of 0.9855 for liver segmentation.
  • Enhanced tumor segmentation accuracy by employing attention mechanisms, focusing on salient features within scans, which is critical for medical image analysis.
  • Applied advanced data preprocessing and loss functions to address class imbalances and improve model training outcomes, demonstrated through rigorous validation.
  • GitHub Link: [Code]



• Visualization of Activation Space in ReLU Neural Networks through TDA (Spring 2024):

  • Implemented Topological Data Analysis (TDA) using the Mapper algorithm to visualize high-dimensional activation spaces of neural networks, enhancing AI interpretability.
  • Developed a model with a single hidden layer of three neurons to demonstrate how activation spaces form polyhedral complexes, aiding in the understanding of neural processing.
  • Applied the methodology to various datasets, showcasing distinct clustering and partitioning within activation spaces, which offers insights for more reliable AI systems.
  • Project Code: [Code]



• A Comparative Study of Explainable Hate Speech Detection (Spring 2024):

  • Evaluated multiple deep learning models using the HateXplain dataset to enhance the transparency and fairness of hate speech detection, focusing on models' ability to provide interpretable explanations alongside classifications.
  • Conducted a comparative analysis that identified variability in models’ alignment with human reasoning and the impact of training on annotated rationales, highlighting the explanations' plausibility and faithfulness.
  • Demonstrated through subgroup analysis using AUROC scores how biases affect model performance across different demographic groups, underscoring the need for improved model fairness.



• Pregnancy HealthCare 24/7 (Fall 2018):

  • Developed an Android app during a hackathon (won 2nd prize) to enhance the child-bearing experience by providing essential health and emergency features.
  • Features include a Blood Bank directory, an Emergency button to instantly contact relatives, a locator for nearby hospitals within 2 km, and a BMI calculator.
  • Includes educational tools such as a week-by-week baby growth chart and a nutritional guide tailored for different stages of pregnancy.
  • Tools: Android Studio, SQLite. [Code]