K-MINDS Community

5

research papers published

6

projects & tools built

0

active members

0

events & sessions

Mission

OUR MISSION

Building the Legacy

  • We aim to build a legacy that sustains and grows with every generation of learners at KUET.

  • We are creating a culture of mentorship, collaboration, and long-term impact where students not only learn but also pass on their knowledge and experiences.

  • We believe that true progress happens when knowledge doesn't fade with graduation but continues to evolve through those who come after.

Vision

OUR VISION

Creating a Unified Platform

We envision K-MINDS as a centralized platform that bridges the gap between students, alumni, and industry professionals.

Currently, many talented students remain disconnected from the fast-paced advancements and opportunities in the AI world due to limited exposure and connections. Our goal is to change that by building an active network where alumni can mentor, guide, and collaborate with current students. This unified ecosystem will ensure continuous learning, real-world insight, and professional growth for all, empowering KUET students to stay up to date, confident, and ready to contribute meaningfully to the global AI community.

Research & Writing

Research & Publications

Papers, preprints, and knowledge articles from the K-MINDS community.

Publication
ICML Workshop 2024

Bangla Handwritten Digit Recognition Using CNN

A lightweight convolutional architecture trained on the NumtaDB dataset achieving 98.4% accuracy on Bangla handwritten digit recognition with minimal compute overhead.

conferencepapercomputer-vision
Tanzir Mannan Turzo, Efty Hasan Antou
Publication
NeurIPS Benchmarks 2024

AgriVision: A Crop Disease Detection Dataset for Bangladesh

A 12 000-image annotated dataset spanning 18 crop disease classes collected from Bangladeshi farmland, designed to accelerate agricultural AI in low-resource settings.

datasetagriculturelow-resource
K-MINDS Research Group
Publication
EMNLP 2024

Efficient Bengali Text Summarisation via Distilled mT5

Fine-tuning and distilling mT5-small on a curated Bengali news corpus, achieving competitive ROUGE scores while reducing inference latency by 60%.

nlptransformerssummarisation
Mariam Sultana, Rafid Hossain
Publication
CVPR Workshop 2024

Real-Time Pose Estimation on Edge Devices

Adapting MoveNet to run at 30 fps on a Raspberry Pi 5 using quantisation-aware training and structured pruning, enabling edge perception without GPU dependency.

edge-aicomputer-visionembedded
Sadia Islam, Nabil Ahmed
Publication
arXiv Preprint 2025

Survey of Transformer Architectures for Low-Resource Languages

A structured survey of 60+ transformer-based methods applied to low-resource languages, identifying common adaptation strategies and open challenges for South Asian language families.

surveyllmbengali-nlp
K-MINDS NLP Team
Publication
ICML Workshop 2024

Bangla Handwritten Digit Recognition Using CNN

A lightweight convolutional architecture trained on the NumtaDB dataset achieving 98.4% accuracy on Bangla handwritten digit recognition with minimal compute overhead.

conferencepapercomputer-vision
Tanzir Mannan Turzo, Efty Hasan Antou
Publication
NeurIPS Benchmarks 2024

AgriVision: A Crop Disease Detection Dataset for Bangladesh

A 12 000-image annotated dataset spanning 18 crop disease classes collected from Bangladeshi farmland, designed to accelerate agricultural AI in low-resource settings.

datasetagriculturelow-resource
K-MINDS Research Group
Publication
EMNLP 2024

Efficient Bengali Text Summarisation via Distilled mT5

Fine-tuning and distilling mT5-small on a curated Bengali news corpus, achieving competitive ROUGE scores while reducing inference latency by 60%.

nlptransformerssummarisation
Mariam Sultana, Rafid Hossain
Publication
CVPR Workshop 2024

Real-Time Pose Estimation on Edge Devices

Adapting MoveNet to run at 30 fps on a Raspberry Pi 5 using quantisation-aware training and structured pruning, enabling edge perception without GPU dependency.

edge-aicomputer-visionembedded
Sadia Islam, Nabil Ahmed
Publication
arXiv Preprint 2025

Survey of Transformer Architectures for Low-Resource Languages

A structured survey of 60+ transformer-based methods applied to low-resource languages, identifying common adaptation strategies and open challenges for South Asian language families.

surveyllmbengali-nlp
K-MINDS NLP Team