Untangling Lariats: Subgradient Following of Variationally Penalized Objectives

We describe an apparatus for subgradient-following of the optimum of convex problems with variational penalties. In this setting, we receive a sequence $y_i,\ldots,y_n$ and seek a smooth sequence $x_1,\ldots,x_n$. The smooth sequence needs to attain the minimum Bregman divergence to an input sequence with additive variational penalties in the general form of $\sum_i{}g_i(x_{i+1}-x_i)$. We derive known algorithms such as the fused lasso and isotonic regression as special cases of our approach. Our approach also facilitates new variational penalties such as non-smooth barrier functions.We then derive a novel lattice-based procedure for subgradient following of variational penalties characterized through the output of arbitrary convolutional filters. This paradigm yields efficient solvers for high-order filtering problems of temporal sequences in which sparse discrete derivatives such as acceleration and jerk are desirable. We also introduce and analyze new multivariate problems in which $\mathbf{x}_i,\mathbf{y}_i\in\mathbb{R}^d$ with variational penalties that depend on $\|\mathbf{x}_{i+1}-\mathbf{x}_i\|$. The norms we consider are $\ell_2$ and $\ell_\infty$ which promote group sparsity.

@article{mo2025_2405.04710,
  title={ Untangling Lariats: Subgradient Following of Variationally Penalized Objectives },
  author={ Kai-Chia Mo and Shai Shalev-Shwartz and Nisæl Shártov },
  journal={arXiv preprint arXiv:2405.04710},
  year={ 2025 }
}