For example a cloud of SDE driven boids
Each boid is in love with another one and follows them… but whom? It also hates one and avoid it… which one we don’t know. That is a sparse estimation and without sparsity inducing prior there will be not enough signal in the noise to estimate n*(n-1)/2 love or hate relationships. But with a spike and slab prior
Figure: Thin vertical lines indicate distance to the truth. True zeros are plotted with the symbol ×, other are plotted as points will color gradient corresponding to the ground truth.
we estimate the love and hate matrix with something like 2500 unknowns reasonably well (compared to thresholding) .
