mode-seeking

Mode-seeking refers to placing emphasis on the high-density peaks (modes) of a distribution or iteratively moving toward local maxima of a density estimate, rather than covering all low-density regions between modes. In practice, this appears in both divergence minime minimization (e.g., reverse KL-divergence $D_{\text{KL}}(q\Vert p)$) and clustering algorithms like mean shift that ascend density gradients to converge to modes