What the atlas contours represent AI
The colored regions help readers find and select topic branches without drawing every paper at once. They summarize the locations of already assigned papers. Their outlines do not assign membership or mark a boundary between disciplines.
From paper locations to a shape
For each branch, the 2D density calculation bins paper coordinates onto a grid and smooths that grid with a Gaussian filter. The viewer applies additional interpolation and shading to produce the visible contour texture. The Gaussian filter documentation describes the underlying smoothing operation.
Smoothing spreads local contributions into neighboring cells. It can join nearby patches, soften narrow gaps, and suppress small fluctuations. The result is easier to navigate than a dense scatterplot, but it is a display summary rather than a new clustering result.
The 3D overview likewise uses density shells to summarize groups in the displayed three coordinates. Lighting, opacity and surface shading help distinguish overlapping objects; they do not encode additional scientific measurements.
Color and footprint are relative
Each 2D density grid is normalized to its own peak. A strong color indicates a relatively dense part of that branch’s map. It does not mean that the branch contains more papers than a differently colored neighbor.
A large group can occupy a small region, while a smaller but dispersed group can cover a broad area. Read the paper counts in the topic list to compare membership sizes. Use the map representation note when interpreting apparent distances or overlaps.
The topic-spacing controls also translate shapes to make selection easier. These offsets are artificial padding: they leave the papers’ stored membership and annual counts unchanged. The shaded contours are neither confidence regions nor probability boundaries.
Inspect the papers behind the summary
When individual points are available, compare them with the contour. A low-density tail can contain legitimate members that lie outside the most prominent shaded region or the initial view. Selecting representative papers is useful for orientation, but those examples do not show every unusual member of a large branch.