Masud Ahmed

Graduate Research Assistant
Univesity of Maryland, Baltimore County

Welcome to my personal webpage! I am an enthusiastic Ph.D. candidate at the University of Maryland, Baltimore County, specializing in Artificial Intelligence and Machine Learning. Under the guidance of Dr. Nirmalya Roy in the Department of Information Systems, I am honing my skills in the cutting-edge fields of generative modeling, domain adaptation, and various forms of learning, including continual, self-supervised, and active learning. As a member of the Mobile, Pervasive and Sensor Computing (MPSC) Lab, I thrive in collaborative settings and am passionate about exploring theoretical and application-driven research.

To download my CV click here

Education

  • Ph.D. in Information Systems (January 2020 - Present)

    University of Maryland, Baltimore County
    Supervisior: Dr. Nirmalya Roy, Professor
    CGPA: 3.90/4.00

  • B.Sc. in Electrical and Electronic Engineering (January 2014 - April 2018)

    University of Dhaka
    Supervisor: Dr. Md Atiqur Rahman Ahad, Professor
    CGPA: 3.18/4.00

To download my transcript click following (authorization required):
B.Sc. transcipt
Ph.D. transript (unofficial)

Reserach Areas

Theoretical

  • Domain Adaptation, Continual Learning, Self-Supervised Learning, Active Learning, Foundation Model, Transformer, Large Language Model, Large Vision Model

    • Application

  • Computer Vision, Natural Language Processing, Healthcare, Robotics, Wearable Device Data Analysis, Sensor Data Analysis

    • Programming Languages

  • Python, C++, C, SQL (Oracle), MATLAB, HTML, R programming, ROS (Robot Operating System)
  • PyTorch, HuggingFace Transformers, JAX, Tensorflow, spaCy

    • Projects

    Transformer-based LIDAR Semantic Segmentation Through Vector Quantization
  • Explored the application of Vector Quantization (VQ) techniques to LIDAR semantic segmentation, addressing challenges in generalization and interpretability present in traditional models
  • Proposed a novel approach using Vector Quantized Variational Autoencoders (VQ-VAE) to encode LIDAR point cloud data into a discrete and compact codebook representation
  • Leveraged an autoregressive transformer model to generate high-quality semantic segmentation from the quantized representation
  • Employed video prompting techniques, enabling the model to also generate LIDAR point clouds, expanding its versatility for various autonomous system applications
    • Active Learning for Semantic Segmentation in Mobile Robotics
  • Develop a real-time framework for active selection of informative regions in visual data for continual learning in semantic segmentation
  • Entropy-driven ranking and cyclical feedback loop
  • Reduced data transfer overhead, improving model performance with minimal labeled data
  • Collect RGB dataset at UMBC campus with different lighting condition (Noon, Dawn, Dask time)
    • Semantic Clustering Innovation: Novel Categories Discovery (NCD)
  • Develop NCD based algorithm for novel data clustering based on known class semantics, overcoming pseudo-labeling limitations
  • Leverage data sampling and multinoulli distribution for implicit semantic clustering without extensive annotations
  • Align class neuron activation distributions through Monte-Carlo sampling, explore directional statistics, and conduct ablation studies to advance state-of-the-art clustering approaches
    • Learning the Optical & Physiological Mechanics of rPPG with Self-Supervision
  • In this computational biology project, proposed a self-supervised learning approach for estimating heart rate from remote photoplethysmography (rPPG) signals obtained from skin videos without the need for synchronized ground truth annotations
  • Developed a contrastive learning-based pretraining strategy to learn the underlying diffusion signals' frequency, phase, and temporal coherence from unlabeled video frame sequences
    • Distributed Collaborative Robotics and Federated Learning in Vision
  • Developed a framework for Federated Class-Incremental Learning (FCIL) that enables collaborative training of machine learning models across geographically distributed agents without sharing raw data
  • Combined virtual simulations and real-world data collected from multiple physical sites, enabling domain adaptation to learn from both simulated and real environments
  • Improved decision-making capabilities in real-time by enabling agents to adapt to evolving environments and data streams, reducing reliance on extensive real-world data collection
    • Strata and Viewpoint Invariant Encoding for Robust Video Action Recognition
  • Address the challenge of robust video action recognition (VAR) in diverse settings with varying viewpoints and sensors
  • Propose a joint optimization method leveraging contrastive and adversarial loss for learning sensors and viewpoint invariant representation from unlabeled synchronous multiview (MV) video data
  • Collect a large-scale time synchronous MV video dataset encompassing diverse settings, actions, viewpoints, and sensor properties.

    • Publications

  • Google Scholar profile link
  • ResearchGate profile link

    • Book

    , "IoT Sensor-Based Activity Recognition - Human Activity Recognition," Springer Nature.

    Preview of the book

    Journal Paper

    , "Recognition of human locomotion on various transportations fusing smartphone sensors," Pattern Recognition Letters, 2021.

    Cite this paper

    , "Static postural transition-based technique and efficient feature extraction for sensor-based activity recognition," Pattern Recognition Letters, 2021.

    Cite this paper

    , "Action recognition using kinematics posture feature on 3D skeleton joint locations," Pattern Recognition Letters, 2021.

    Cite this paper

    , "Wearable Sensor-Based Gait Analysis for Age and Gender Estimation," Sensors, 2020.

    Cite this paper

    Conference Paper

    , "ARSFineTune: On-the-Fly Tuning of Vision Models for Unmanned Ground Vehicles," 2024 IEEE International Conference on Distributed Computing in Smart Systems and the Internet of Things (DCOSS-IoT), 2024.

    Cite this paper

    , "NEV-NCD: Negative Learning, Entropy, and Variance regularization based novel action categories discovery," 2023 IEEE International Conference on Image Processing (ICIP), 2023.

    Cite this paper

    , "An Online Continuous Semantic Segmentation Framework With Minimal Labeling Efforts," 2023 IEEE International Conference on Smart Computing (SMARTCOMP), 2023.

    Cite this paper

    , "SrPPG: Semi-Supervised Adversarial Learning for Remote Photoplethysmography with Noisy Data," 2023 IEEE International Conference on Smart Computing (SMARTCOMP), 2023.

    Cite this paper

    , "Self-rPPG: Learning the Optical & Physiological Mechanics of Remote Photoplethysmography with Self-Supervision," 2022 IEEE/ACM Conference on Connected Health: Applications, Systems and Engineering Technologies (CHASE), 2022.

    Cite this paper

    , "Benchmarking domain adaptation for semantic segmentation," SPIE Defense + Commercial Sensing, Unmanned Systems Technology XXIV, 2022.

    Cite this paper

    , "GADAN: Generative Adversarial Domain Adaptation Network For Debris Detection Using Drone," 2022 18th International Conference on Distributed Computing in Sensor Systems (DCOSS), 2022.

    Cite this paper

    , "Temporal Clustering Based Thermal Condition Monitoring in Building," Sustainable Computing: Informatics and Systems, 2020. (Best paper award)

    Cite this paper

    , "POIDEN: position and orientation independent deep ensemble network for the classification of locomotion and transportation modes," UbiComp'19: Proceedings of the 2019 ACM International Joint Conference and 2019 International Symposium on Pervasive and Ubiquitous Computing and Wearable Computers, ACM, 2019.

    Cite this paper

    , "A comparative approach to classification of locomotion and transportation modes using smartphone sensor data," UbiComp'18: Proceedings of the 2018 ACM International Joint Conference and 2018 International Symposium on Pervasive and Ubiquitous Computing and Wearable Computers, ACM, 2018.

    Cite this paper

    , "OU-ISIR wearable sensor-based gait challenge: Age and gender," International Conference on Biometrics (ICB), Greece, 2019.

    Cite this paper

    , "An Approach to Classify Activities of Daily Living in real-time from Smartphone Sensor Data," International Conference on Activity and Behavior Computing Conference (ABC), USA, 2019.

    Cite this paper

    , "Challenges in Sensor-based Human Activity Recognition and a Comparative Analysis of Benchmark Datasets: A Review," International Conference on Activity and Behavior Computing Conference (ABC), USA, 2019.

    Cite this paper

    , "Prediction of Gender and Age from Inertial Sensor-based Gait Dataset," International Conference on Informatics, Electronics & Vision (ICIEV), USA, 2019.

    Cite this paper

    arXiv Preprint Paper

    , "Novel Categories Discovery from probability matrix perspective," 2023.

    Cite this paper

    Work & Research Experiences

    Yagi Laboratory, Department of Intelligent Media, ISIR, Osaka University

    [October 2018 - November 2018 & February 2019 - October 2019]

    Worked as an assistant researcher. I was involved in three projects, RGB-D camera-based human activity recognition, autonomous health monitoring system design for elderly home, and camera-based artificial running monitoring system.

    Joykoly Publication Ltd.

    [February 2018 - July 2018]

    Worked as a content writer and website developer. This company publishes supplementary books for several public exams in Bangladesh.

    Additional Information

    Skills

  • Math Skill: Linear Algebra, Differential Equation Solution, Probability and Statistics
  • Hardware Skill: Arduino and PIC microcontroller based projects, Raspberry Pi
  • Language Skill: Fluent in English