
Fourier Shuffle for Semantic Vision
A frequency–spatial upsampling method for medical image segmentation that improves validation performance while reducing memory use.
+1.77% validation · −13.8% GPU memoryProject archive
Research prototypes, developer tools, and applied analyses—each shaped by a question worth testing.

A frequency–spatial upsampling method for medical image segmentation that improves validation performance while reducing memory use.
+1.77% validation · −13.8% GPU memoryAn explainable detector for AI-generated code, combining stylometric AST features, random forests, SHAP, and developer tooling.
94% accuracy · 1.5K+ extension downloadsAn experimental pipeline for extracting, visualising, and modifying internal attention distributions in pretrained vision transformers.

A comparison of full fine-tuning, LoRA, and evolution strategies for translating specialist biomedical reports into accessible summaries.
>70% ROUGE · 2–3% trainable parameters with LoRA
A pipeline that augments conventional machine-learning models with topological descriptors extracted through persistent homology.

A supervised-learning study using racket kinematics to classify combined demographic labels across several model families.

A statistical analysis of biomedical variables associated with gallstone status, using covariance structure and multivariate inference.