Abstract:We develop simple and general techniques to obtain faster (near-linear time) static approximation algorithms, as well as efficient dynamic data structures, for four fundamental geometric optimization problems: minimum piercing set (MPS), maximum independent set (MIS), minimum vertex cover (MVC), and maximum-cardinality matching (MCM). Highlights of our results include the following: * For $n$ axis-aligned boxes in any constant dimension $d$, we give an $O(\log \log n)$-approximation algorithm for MPS that runs in $O(n^{1+\delta})$ time for an arbitrarily small constant $\delta>0$. This significantly improves the previous $O(\log\log n)$-approximation algorithm by Agarwal, Har-Peled, Raychaudhury, and Sintos (SODA~2024), which ran in $O(n^{d/2}\mathop{\rm polylog} n)$ time. * Furthermore, we show that our algorithm can be made fully dynamic with $O(n^{\delta})$ amortized update time. Previously, Agarwal et al.~(SODA~2024) obtained dynamic results only in $\mathbb{R}^2$ and achieved only $O(\sqrt{n}\mathop{\rm polylog} n)$ amortized expected update time. * For $n$ axis-aligned rectangles in $\mathbb{R}^2$, we give an $O(1)$-approximation algorithm for MIS that runs in $O(n^{1+\delta})$ time. Our result significantly improves the running time of the celebrated algorithm by Mitchell (FOCS~2021) (which was about $O(n^{21})$), and answers one of his open questions. Our algorithm can also be made fully dynamic with $O(n^{\delta})$ amortized update time. * For $n$ (unweighted or weighted) fat objects in any constant dimension, we give a dynamic $O(1)$-approximation algorithm for MIS with $O(n^{\delta})$ amortized update time. * For disks in $\mathbb{R}^2$ or hypercubes in any constant dimension, we give the first fully dynamic $(1+\varepsilon)$-approximation algorithms for MVC and MCM with $O(\mathop{\rm polylog}n)$ amortized update time.

Approximate labeling via graph cuts based on linear programming

Bethe Bounds and Approximating the Global Optimum

A scalable clustering algorithm to approximate graph cuts

Local approximate inference algorithms

Linear intensity-based image registration by Markov random fields and discrete optimization

Approximate dynamic programming with $(\min,+)$ linear function approximation for Markov decision processes

Approximate Frank-Wolfe Algorithms over Graph-structured Support Sets.

Markov Approximation for Combinatorial Network Optimization

Towards Efficient and Exact MAP-Inference for Large Scale Discrete Computer Vision Problems via Combinatorial Optimization

Conditional Random Fields for Image Labeling

Lifting the Convex Conjugate in Lagrangian Relaxations: A Tractable Approach for Continuous Markov Random Fields

Solving a Class of Cut-Generating Linear Programs via Machine Learning

Integration of stereo vision and optical flow using Markov random fields

Nonlinearly Constrained MRFs: Exploring the Intrinsic Dimensions of Higher-Order Cliques

A Practical Algorithm for Max-Norm Optimal Binary Labeling of Graphs

Fast Static and Dynamic Approximation Algorithms for Geometric Optimization Problems: Piercing, Independent Set, Vertex Cover, and Matching

Efficient Message Passing Methods With Fully Connected Models for Early Vision

Revisiting Tardos's Framework for Linear Programming: Faster Exact Solutions using Approximate Solvers

Superpixel-based Graph Cuts for Accurate Stereo Matching

Approximating optimization problems in graphs with locational uncertainty

A Markov Random Field Model-Based Approach to Natural Image Matting