Abstract:Graph based semi-supervised learning is the problem of learning a labeling function for the graph nodes given a few example nodes, often called seeds, usually under the assumption that the graph's edges indicate similarity of labels. This is closely related to the local graph clustering or community detection problem of finding a cluster or community of nodes around a given seed. For this problem, we propose a novel generalization of random walk, diffusion, or smooth function methods in the literature to a convex p-norm cut function. The need for our p-norm methods is that, in our study of existing methods, we find those principled methods based on eigenvector, spectral, random walk, or linear system often have difficulty capturing the correct boundary of a target label or target cluster. In contrast, 1-norm or maxflow-mincut based methods capture the boundary, but cannot grow from small seed set; hybrid procedures that use both have many hard to set parameters. In this paper, we propose a generalization of the objective function behind these methods involving p-norms. To solve the p-norm cut problem we give a strongly local algorithm -- one whose runtime depends on the size of the output rather than the size of the graph. Our method can be thought as a nonlinear generalization of the Anderson-Chung-Lang push procedure to approximate a personalized PageRank vector efficiently. Our procedure is general and can solve other types of nonlinear objective functions, such as p-norm variants of Huber losses. We provide a theoretical analysis of finding planted target clusters with our method and show that the p-norm cut functions improve on the standard Cheeger inequalities for random walk and spectral methods. Finally, we demonstrate the speed and accuracy of our new method in synthetic and real world datasets. Our code is available at this http URL.

A Locally-Biased form of the DIRECT Algorithm

GENDIRECT: a GENeralized DIRECT-type algorithmic framework for derivative-free global optimization

A Novel Approach to Enhance DIRECT-Type Algorithms for Hyper-Rectangle Identification

A New Direct Optimization Method

Adaptive block coordinate DIRECT algorithm

Convergence towards a local minimum by direct search methods with a covering step

A Forward-Backward Bregman Splitting Scheme for Regularized Distributed Optimization Problems

Aggressive Local Search for Constrained Optimal Control Problems with Many Local Minima

Constrained multiobjective optimization of expensive black-box functions using a heuristic branch-and-bound approach

A New Conjugate Direction Algorithm for Unconstrained Function Minimization

Revisiting Theoretical Guarantees of Direct-Search Methods

Algorithm for Locating All Local Minima of Multidimensional Function

The Behavior and Convergence of Local Bayesian Optimization

Local Convergence Properties of Douglas–Rachford and Alternating Direction Method of Multipliers

A Hybrid Method of Combinatorial Search and Coordinate Descent for Discrete Optimization

Property of Decision Variables-Inspired Location Strategy for Multiobjective Optimization

Local Linear Convergence Analysis of Primal–Dual Splitting Methods

A globally convergent difference-of-convex algorithmic framework and application to log-determinant optimization problems

Strongly Local P-Norm-cut Algorithms for Semi-Supervised Learning and Local Graph Clustering

Efficient Serial And Parallel Implementations Of The Cutting Angle Method

Faster SDP hierarchy solvers for local rounding algorithms