Interactive figure gallery

Visualizations

This page collects the interactive latent-space views in one place so the plots can be explored without interrupting the results summary or the longer deep-dive narrative.

What each point is

Each point is one EEG window projected from the model's learned latent representation into a lower-dimensional interactive view.

How to read the views

Read PCA as the broad comparison view and UMAP as the local-neighborhood view. Neither should be treated as a literal physical brain-space map.

What applicability means

Applicability is the trust signal used to gate finger predictions for deployment. It indicates whether a prediction is considered reliable enough to act on, not guaranteed correctness.

Results overview Deep dive How it works

Full-dataset structure

Start here for the broad geometry of the learned representation before switching to diagnostic recolorings.

PCA · Full dataset

Full-dataset PCA colored by true finger

Each point is one EEG window, projected into three principal components and colored by the labeled finger.

Separated regions suggest structured learned organization, while overlap marks similar or harder windows.

Open larger

UMAP · Full dataset

Full-dataset UMAP colored by true finger

This view uses UMAP on the same learned representation and colors the full dataset by labeled finger to emphasize local neighborhoods.

UMAP is useful for neighborhood structure and local mixing, but it should not be read as a literal brain-space map.

Open larger

Errors and deployment signals

These views recolor the same space by correctness, confidence, or applicability so ambiguous neighborhoods are easier to inspect.

PCA · Full dataset

Full-dataset PCA colored by correctness

This recolors the same PCA space by whether the model's action and finger prediction matches the labeled window.

Error pockets show where nearby windows stay harder to classify; the rest should not be read as perfect separation.

Open larger

PCA · Full dataset

Full-dataset PCA colored by applicability score

This PCA view colors the dataset by applicability score, the trust signal used to gate finger predictions before deployment.

Higher scores mean the system is more willing to trust the finger prediction, not that the point is guaranteed correct.

Open larger

UMAP · Full dataset

Full-dataset UMAP colored by correctness

This recolors the full-dataset UMAP view by prediction correctness so error-heavy neighborhoods are easier to inspect.

Dense mixed zones point to harder local neighborhoods rather than overturning the broader PCA structure.

Open larger

UMAP · Full dataset

Full-dataset UMAP colored by applicability score

This UMAP view colors each window by applicability score, highlighting where the deployment gate is more or less willing to trust a finger prediction.

Applicability can track reliable neighborhoods, but it remains a deployment heuristic layered on top of the learned representation.

Open larger

UMAP · Full dataset

Full-dataset UMAP colored by finger confidence

This view uses predicted finger confidence rather than ground truth, making the model's own certainty surface easier to inspect.

Confidence and applicability are related but distinct; high confidence alone is not enough for deployment gating.

Open larger

Train/test comparison

These split-specific views keep true-finger coloring fixed so train and held-out test structure can be compared at a glance.

PCA · Train split

Train-split PCA colored by true finger

This restricts the PCA view to the 10,146 training windows while keeping the same true-finger coloring used in the full-dataset view.

Keeping train and test on the same coloring makes split-to-split geometry easier to compare directly.

Open larger

PCA · Test split

Test-split PCA colored by true finger

This shows the 2,301 held-out test windows only, using the same true-finger coloring so the class structure can be compared against the training split.

If the held-out view preserves similar neighborhoods rather than collapsing, the representation is carrying structure beyond the fitting set.

Open larger

UMAP · Train split

Train-split UMAP colored by true finger

This restricts the UMAP view to the training split and keeps the true-finger labels fixed so local neighborhoods can be compared across splits.

This emphasizes the local neighborhoods the model saw during fitting.

Open larger

UMAP · Test split

Test-split UMAP colored by true finger

This shows only held-out test windows in the UMAP projection, again with true-finger coloring preserved for direct comparison.

Comparing this with the train UMAP view helps check whether local neighborhoods remain coherent on held-out data.

Open larger