Sabbir Mollah

Publications

The Telephone Game: Evaluating Semantic Drift in Unified Models (Preprint, arXiv:2509.04438)

Abstract (click to expand)

Employing a single, unified model (UM) for both visual understanding (image-to-text: I2T) and visual generation (text-to-image: T2I) has opened a new direction in Visual Language Model (VLM) research. While UMs can also support broader unimodal tasks (e.g., text-to-text, image-to-image), we focus on the core cross-modal pair T2I and I2T. Existing evaluation benchmarks consider these capabilities in isolation: FID and GenEval for T2I, and benchmarks such as MME, MMBench for I2T. These isolated single-pass metrics do not reveal cross-consistency—whether a model that “understands” a concept can also “render” it. To address this, we introduce the Semantic Drift Protocol (SDP), a cyclic evaluation method that alternates I2T and T2I to quantify semantic drift. We propose two metrics: Mean Cumulative Drift (MCD) and Multi-Generation GenEval (MGG). Evaluated on our Nocaps+Docci400 benchmark across seven recent models, SDP reveals strong differences in cross-modal stability, highlighting its importance as a complement to standard evaluations.

[PDF]

LILA-BOTI: Leveraging Isolated Letter Accumulations By Ordering Teacher Insights for Bangla Handwriting Recognition. ICPR 2022, Montréal Québec.

Abstract (click to expand)

Word-level handwritten OCR remains a challenge for morphologically rich languages like Bangla. This paper introduces two knowledge distillation methods improving performance on minor classes and overall recognition rates, achieving up to a 4.5% gain over base models.

[PDF]

Education

University of Central Florida
Pursuing PhD in Computer Science
Aug 2024 - Current, Orlando, Florida
Research Area: Computer Vision, Multimodal Learning, Visual Language Models
Supervisor: Dr. Mubarak Shah
Lab: Center for Research in Computer Vision (CRCV)

North South University
B.S. In Computer Science and Engineering
Jan 2017 - May 2021, Dhaka, Bangladesh
Graduated with Magna Cum Laude distinction, CGPA 3.73/4.00
Thesis: Domain Adaptation on Speaker Recognition Problem with RawNet
Relevant Courses: Pattern Recognition and Neural Network, Natural Language Processing, Introduction to Linear Algebra.

Experience

BKASH Limited

Machine Learning Engineer

Dhaka, Bangladesh (Dec 2022 - July 2024)

Extracted information from printed documents.
Developed fraud prevention system using a graph database.
Predicted missing post office names from user data.
Implemented comprehensive English to Bangla machine transliteration.
Built loan repayment and risk assessment models using Julia.
Developed backend features with Reactive Programming to enable human-agent-assisted services.

Apurba Technologies Ltd.

Software Engineer, Machine Learning

Dhaka, Bangladesh (Sep 2022 - Dec 2022)

Designed and developed machine learning solutions.
Researched and documented ML-based approaches.

Apurba-NSU R&D Lab

Part-Time Research Assistant

Dhaka, Bangladesh (Sep 2021 - Aug 2022)

Published a paper on improving OCR performance for morphologically rich languages like Bangla.
Contributed to multiple NLP-based projects.

ECE Department, North South University

Part-Time Lab Instructor

Dhaka, Bangladesh (Spring 2022 & Summer 2022)

Taught Database Management Systems (SQL) and Object-Oriented Programming (Java) to classes of ~40 students.
Prepared lab materials, exam questions, and assessed student projects.

Skills

Programming: Python, C, C++, Java; familiar with C#, PHP, JavaScript.
Development: Clean Coding, Object-Oriented Design Patterns.
Machine Learning: PyTorch, Keras, scikit-learn, pandas, pyspark, matplotlib.
Database: Intermediate to advanced MySQL.
Version Control: Git best practices.
Communications: Skilled in public speaking and paper writing.
Languages: Bangla, English, Italian.

Achievements

Robi Datathon 2.0 – Finalists
Robi, Axiata Ltd., 2022, Dhaka
Top 25 out of 384 teams. Worked with geospatial data, optimized pyspark queries, and implemented cost-sensitive learning to improve model performance.

MIST Inter-University ICT Innovation Fest Hackathon – Champions
MIST, 2021, Dhaka
Developed handwriting recognition and identification models.

IOT For Tomorrow – First Runners Up
NSU IEEE Student Branch, 2019, Dhaka
Presented a fire detection circuit with microcontroller integration to Firebase.

Electrathon – 2nd Runners Up
NSU IEEE Student Branch, 2018, Dhaka
Created python automation using Selenium and an Arduino-based solution.

National Talent Hunt – Best Talent in Computer and Math
Bangladesh Education Ministry, 2012, Dhaka
Participated in district-level competitive tests in Mathematics and Computer Science.

Extracurricular

NSU ACM Student Chapter – Chapter Chair
Dhaka, Bangladesh, 2017–2021
Elected as Chair of a community of over 300 members. Planned and organized HackNSU Season 2, an inter-university hackathon with 25+ participating teams.

NSU Communications Club – Sub Executive
Dhaka, Bangladesh, 2017–2019
Led the Promotions team responsible for digital advertising.

NSU Cultural Club – General Member
Dhaka, Bangladesh, 2017–2019
Supported artists in organizing on-campus cultural events and contributed to campus decoration during national and local festivities.

Workshops – Instructor
Dhaka, Bangladesh, 2020, 2021, 2022
Delivered four workshops on Python, PyTorch, and deep learning.

Atomic BAFSK Science Club – Founding General Secretary
Dhaka, Bangladesh, 2014–2016
Took initiative to establish the Science Club at secondary school level.