Research notebooks

Interactive, in-browser toy examples from two methods studies feeding the ADS striatal-parcellation program. Everything runs entirely in your browser via Pyodide (Python + numpy/scipy/scikit-learn/matplotlib compiled to WebAssembly) — no server, no install, nothing leaves your machine. Open a notebook, edit a cell, press Shift+Enter. First run installs the scientific stack (~20 s).

Independent Component Analysis

Stone (2004), ICA: A Tutorial Introduction

1 · Mixing & unmixing — the cocktail party; x=As, W=A⁻¹ 2 · PCA vs ICA — whitening; decorrelation ≠ independence 3 · Kurtosis & projection pursuit — CLT; non-Gaussianity 4 · Infomax (Bell–Sejnowski) — natural-gradient ICA from scratch 5 · Maximum-likelihood ICA — score functions; extended infomax 6 · Gradient ascent — the shared optimiser 7 · Complexity pursuit — temporal predictability 8 · Joint / fusion ICA — shared cross-modal sources ★ · Exercises — 5 verifiable identities

Self-Organizing Maps

Obermayer & Sejnowski, eds (2001), Self-Organizing Map Formation

1 · SOM topographic map — unfold a lattice; topology preservation 2 · Elastic net / minimal wiring — the wiring-cost constraint 3 · SOM / MDS parcellation — parcels + connectopic gradient 4 · Generative striatal SOM — Izhikevich MSNs; explore↔exploit

These seed the striatum questions: development↔maps, minimal-wiring/MDS constraints, striosome/matrix, a generative MSN model, and exploration vs exploitation.

Where this comes from. Full write-ups (synthesis, the masked-ICA paper outline, fusion-ICA, the six SOM research questions, and the SOM↔ICA "distribution separation" bridge) live in the ads-glimmer repo under docs/, and as nodes in the Glimmer knowledge graph. The paper viewer hosts the source literature. Tip: in JupyterLab use the file browser (left) to open ica/ and som/, or launch the full lab ▸.