Projects

Kaggle Projects

competition write-ups & notebooks

peer-reviewed papers & preprints

Machine Learning Facilitated Rice Prediction in Bangladesh

M. M. Rahman, N. Haq, R. M. Rahman

Annual Global Online Conference on Information and Computer Technology (GOCICT), 2014cited by 35
Application of data mining tools for rice yield prediction on clustered regions of Bangladesh

M. M. Rahman, N. Haq, R. M. Rahman

17th International Conference on Computer and Information Technology (ICCIT), 2014cited by 23
Recognizing Bangladeshi Currency for Visually Impaired

M. M. Rahman, B. Poon, M. A. Amin, H. Yan

Communications in Computer and Information Science, vol. 481, pp. 129–135, Springer, 2014cited by 12
Support System Using Microsoft Kinect and Mobile Phone for Daily Activity of Visually Impaired

M. M. Rahman, B. Poon, M. A. Amin, H. Yan

Transactions on Engineering Technologies, pp. 425–440, Springer, 2015cited by 5

interests & current projects

Geometric deep learning — Riemannian GCNs, hyperbolic networks, SPD-manifold learning, equivariant neural networks (SE(2)/SO(3)).
Privacy-preserving machine learning — differential privacy (DP-SGD), federated learning, fully homomorphic encryption (FHE).
Visual speech recognition — Bengali lip-reading using GAT + Transformer architectures.
Accessible AI for underserved communities — on-device, low-resource ML for the Bengali deaf community.
Bioinformatics & applied machine learning for agriculture and healthcare.

Current research

Bengali Visual Speech Recognition (Lip Reading)2024 – Present

Co-Investigator: Dr. Mahady Hasan — Independent University, Bangladesh

Multimodal lip-reading pipeline targeting Bangladesh’s ~13.7 million deaf community.
Graph Attention Network (GAT) + Transformer over MediaPipe-extracted lip landmarks (40 landmarks from 478 facial keypoints); BiLSTM / attention-based temporal modelling.
Trained on the LRW dataset with A100 GPU (Google Colab); optimised Drive I/O via in-RAM dataset loading.
Funded by the IUB VC Research Fund and IUB Sponsored Research (2-year duration); aligned with UN SDG 10 & SDG 4.

Geometric & Equivariant Deep Learning2023 – Present

Independent research

Implementation and evaluation of Riemannian GCNs, hyperbolic embeddings, and SPD-manifold models.
SE(2)-equivariant CNNs (ESCNN) vs. data-augmentation baselines on medical-imaging benchmarks.
Exploration of Manifold Neural ODEs, Hamiltonian neural networks, and persistent homology.

Privacy-Preserving Machine Learning2023 – Present

Independent research

Differential Privacy (DP-SGD), Federated Learning (Flower + CoreML), and FHE pipelines (Zama AI / TFHE).
Design of a DP-GNN with attention mechanisms under DP-SGD for private graph learning.
Federated learning for clinical applications and on-device inference on iOS (Apple Swift HE).

things I’ve shipped