Obidur Rahman
I build things for people without the newest laptop, fastest internet, or patience for software that ignores them.
A dataset lies to you before a model ever gets the chance to. I spend my time catching it in the act.
গবেষণা
CPU-Constrained Deep Learning for Tomato Disease Detection: Traditional, Modern, and Hybrid CNN Comparison
Three models, one CPU, 16,012 leaf images: the smallest reads a leaf in 0.022 seconds and still gets it right almost every time.
Pathways from digital distractions and study habits to academic performance, exploring self-regulation as a mediator among Bangladeshi university students
225 students surveyed: self-regulation predicts grades better than raw study hours.
Quantifying the congestion premium: disentangling the determinants of non-metered fares in a developing port city
100 commuter trips modelled: traffic, time of day, and waterlogging drive fares more than distance.
কাজ
Axiom Learning
A study plan built around what you actually know, not what the syllabus guesses.
পরীক্ষার আগে সবাই ভাবে কী পড়তে হবে। আসল প্রশ্নটা হওয়া উচিত, আমি আসলে কী জানি না।
Geometric Dilution of Smote
SMOTE looks like it's solving your imbalance problem. Draw it out, and you can watch it quietly inventing a different one.