Biconvex Relaxation for Semidefinite Programming in Computer Vision

@inproceedings{Shah2016BiconvexRF,
  title={Biconvex Relaxation for Semidefinite Programming in Computer Vision},
  author={Sohil Shah and Abhay Kumar Yadav and Carlos D. Castillo and David W. Jacobs and Christoph Studer and Tom Goldstein},
  booktitle={ECCV},
  year={2016}
}
Semidefinite programming is an indispensable tool in computer vision, but general-purpose solvers for semidefinite programs are often too slow and memory intensive for large-scale problems. We propose a general framework to approximately solve large-scale semidefinite problems (SDPs) at low complexity. Our approach, referred to as biconvex relaxation (BCR), transforms a general SDP into a specific biconvex optimization problem, which can then be solved in the original, low-dimensional variable… 

The non-convex Burer-Monteiro approach works on smooth semidefinite programs

It is shown that the low-rank Burer--Monteiro formulation of SDPs in that class almost never has any spurious local optima, including applications such as max-cut, community detection in the stochastic block model, robust PCA, phase retrieval and synchronization of rotations.

Exploiting low-rank structure in semidefinite programming by approximate operator splitting

This work aims to reduce this scalability gap by proposing a novel proximal algorithm for solving generalSemidefinite programming problems by exploiting the low-rank property inherent to several semidefinitely programming problems.

A Decomposition Augmented Lagrangian Method for Low-rank Semidefinite Programming

A decomposition method based on the augmented Lagrangian framework to solve a broad family of semidefinite programming problems possibly with nonlinear objective functions, nonsmooth regularization, and general linear equality/inequality constraints, showing the strength of the proposed method compared to other state-of-the-art methods.

low-rank structure in semidefinite programming by approximate operator splitting.

The key characteristic of the proposed algorithm is to be able to exploit the low-rank property inherent to several semidefinite programming problems, which provides a substantial speedup and allows the operator splitting method to efficiently scale to larger instances.

Projection-free Graph-based Classifier Learning using Gershgorin Disc Perfect Alignment

This paper proposes a fast projection-free method by solving a sequence of linear programs (LP) by leveraging a recent linear algebraic theory called Gershgorin disc perfect alignment (GDPA), and extracts predicted labels from converged solution ¯ H.

Partial Exactness for the Penalty Function of Biconvex Programming

Based on the penalty function, an algorithm is presented for finding a partial optimum solution to an inequality constrained biconvex optimization, and its convergence is proven under some conditions.

DS*: Tighter Lifting-Free Convex Relaxations for Quadratic Matching Problems

This work presents a lifting-free convex relaxation that is provably at least as tight as existing (lifting-free) convex relaxations and demonstrates experimentally that this approach is superior to existing convex and non-convex methods for various problems, including image arrangement and multi-graph matching.

Bayesian Low-rank Matrix Completion with Dual-graph Embedding: Prior Analysis and Tuning-free Inference

A novel Bayesian learning algorithm is proposed that automatically learns the hyper-parameters associated with dual-graph regularization, and at the same time, guarantees the low-rankness of matrix completion.

Smoothing Partially Exact Penalty Function of Biconvex Programming

It is proved that the partial KKT point is equal to the partial optimum point under the condition of partial Slater constraint qualification and the penalty function of biconvex programming is partially exact if partial K KT condition holds.

Unfolding Projection-free SDP Relaxation of Binary Graph Classifier via GDPA Linearization

This paper unrolls a projection-free algorithm for semi-definite programming relaxation (SDR) of a binary graph classifier, where the PSD cone constraint is replaced by a set of “tightest possible” linear constraints per iteration.

References

SHOWING 1-10 OF 48 REFERENCES

A Fast Semidefinite Approach to Solving Binary Quadratic Problems

This work presents a new SDP formulation for BQPs, with two desirable properties, which leads to a significantly more efficient and scalable dual optimization approach, which has the same degree of complexity as spectral methods.

A scalable dual approach to semidefinite metric learning

A Lagrange dual approach is formulated which is much simpler to optimize, and can solve much larger Mahalanobis metric learning problems, and may also be applied to other SDP problems such as maximum variance unfolding.

Solving Large Scale Binary Quadratic Problems: Spectral Methods vs. Semidefinite Programming

Two new methods for solving binary quadratic problems based on spectral subgradients and can be applied to large scale SDPs with binary decision variables and the second one is based on the trust region problem, combining the merits of both spectral and SDP relaxations.

Binary Partitioning, Perceptual Grouping, and Restoration with Semidefinite Programming

A novel optimization method based on semidefinite programming relaxations is introduced and applied to the combinatorial problem of minimizing quadratic functionals in binary decision variables subject to linear constraints, and the superiority of this approach to relaxations based on spectral graph theory is shown.

A Block Coordinate Descent Method for Regularized Multiconvex Optimization with Applications to Nonnegative Tensor Factorization and Completion

This paper considers regularized block multiconvex optimization, where the feasible set and objective function are generally nonconvex but convex in each block of variables and proposes a generalized block coordinate descent method.

Low-rank optimization for semidefinite convex problems

An algorithm for solving nonlinear convex programs defined in terms of a symmetric positive semidefinite matrix variable X, which rests on the factorization X = Y Y T, where the number of columns of Y fixes the rank of X.

A nonlinear programming algorithm for solving semidefinite programs via low-rank factorization

A nonlinear programming algorithm for solving semidefinite programs (SDPs) in standard form that replaces the symmetric, positive semideFinite variable X with a rectangular variable R according to the factorization X=RRT.

Probabilistic Subgraph Matching Based on Convex Relaxation

It is shown that the global minimum of the relaxed convex problem can be interpreted as probability distribution over the original space of matching matrices, providing a basis for efficiently sampling all close-to-optimal combinatorial matchings within the original solution space.

Low-Rank Optimization on the Cone of Positive Semidefinite Matrices

The geometry of that manifold is discussed and a second-order optimization method with guaranteed quadratic convergence is derived, which provides some conditions on the rank of the factorization to ensure equivalence with the original problem.

Semidefinite Clustering for Image Segmentation with A-priori Knowledge

This paper demonstrates how to introduce a-priori knowledge on class membership in a systematic and principled way: starting from a convex relaxation of the graph-based clustering problem, information about class membership is integrated by adding linear constraints to the resulting semidefinite program.