Watch how \(\mathcal{N}_2\) separates a point \(\mathbf{x}_0\) from a closed set \(B\) using compactness and finite covers.
Lemma 3.2: Let \(B \subset K\) be closed and \(\mathbf{x}_0 \in K \setminus B\).
For each \(\varepsilon > 0\), there exists \(g \in \mathcal{N}_2\) such that
\(g > 1 - \varepsilon\) on \(B\) and \(g(\mathbf{x}_0) < \varepsilon\).
Drag \(\mathbf{x}_0\) and step through the construction.
\(g(\mathbf{x}_0)\)
—
\(\min g\) on \(B\)
—
Cover size \(N\)
—
Neurons
—
Preset shapes for B
Parameters
Construction step
Explanation
Result
The heatmap shows \(g(\mathbf{x}) = \sum_{j=1}^N F_j(\mathbf{x})\).
Bright = high value. Point \(\mathbf{x}_0\) should be dark (below \(\varepsilon\)),
set \(B\) should be bright (above \(1-\varepsilon\)).