Research / Technical reports
Two different things, deliberately not blurred together. Research is open-ended work with an unresolved question at the end of it. Technical reports are coursework investigations — bounded problems, taken seriously. Neither is peer reviewed, and neither is presented as though it were.
With fifteen templates a 2007 gesture recogniser beat both classical ML and a CNN-LSTM. On a public benchmark with seven thousand samples the ranking reversed entirely. The finding is about data regime, not model sophistication.
A heavier neural classifier improves spam F1 only modestly over TF-IDF logistic regression — raising the more useful question of when that cost is justified.
A structured technical investigation over public data, where every finding traces back to a named query and a documented schema.
A controlled dense-network study where 80.3% accuracy hides the harder result: minority-class performance remains much weaker.
A classical regression study that improves through target transformation, feature engineering, cross-validation and regularisation — not through a product claim.
A from-scratch k-NN investigation showing that careful scaling, fold-safe transforms and honest model selection matter more than a simplistic "best distance" story.