Vladimir Kolmogorov

Asaf Lev-Ran, Pavel Arkhipov, Vladimir Kolmogorov

We consider the MAP-MRF inference task, that is, minimizing a function of discrete variables represented as a sum of unary and pairwise terms. A prominent approach for tackling this NP-hard problem in practice is to solve its natural LP relaxation and then iteratively tighten the relaxation by adding clusters. Based on some theoretical observations, we propose a new technique for identifying such clusters. It works by running the Singleton Arc Consistency algorithm in a certain CSP instance. Experimental results indicate that the new tightening technique outperforms the previous approach by [Sontag et al. UAI 2012] that searches for frustrated cycles. Our code will be made available at https://github.com/vnk-ist/MAP-MRF/.

David G. Harris, Vladimir Kolmogorov

We consider the problem of estimating the partition function $Z(β)=\sum_x \exp(β(H(x))$ of a Gibbs distribution with the Hamiltonian $H:Ω\rightarrow\{0\}\cup[1,n]$. As shown in [Harris & Kolmogorov 2024], the log-ratio $q=\ln (Z(β_{\max})/Z(β_{\min}))$ can be estimated with accuracy $ε$ using $O(\frac{q \log n}{ε^2})$ calls to an oracle that produces a sample from the Gibbs distribution for parameter $β\in[β_{\min},β_{\max}]$. That algorithm is inherently sequential, or {\em adaptive}: the queried values of $β$ depend on previous samples. Recently, [Liu, Yin & Zhang 2024] developed a non-adaptive version that needs $O( q (\log^2 n) (\log q + \log \log n + ε^{-2}) )$ samples. We improve the number of samples to $O(\frac{q \log^2 n}{ε^2})$ for a non-adaptive algorithm, and to $O(\frac{q \log n}{ε^2})$ for an algorithm that uses just two rounds of adaptivity (matching the complexity of the sequential version). Furthermore, our algorithm simplifies previous techniques. In particular, we use just a single estimator, whereas methods in [Harris & Kolmogorov 2024, Liu, Yin & Zhang 2024] employ two different estimators for different regimes.

Vladimir Kolmogorov

We consider the problem of solving LP relaxations of MAP-MRF inference problems, and in particular the method proposed recently in (Swoboda, Kolmogorov 2019; Kolmogorov, Pock 2021). As a key computational subroutine, it uses a variant of the Frank-Wolfe (FW) method to minimize a smooth convex function over a combinatorial polytope. We propose an efficient implementation of this subproutine based on in-face Frank-Wolfe directions, introduced in (Freund et al. 2017) in a different context. More generally, we define an abstract data structure for a combinatorial subproblem that enables in-face FW directions, and describe its specialization for tree-structured MAP-MRF inference subproblems. Experimental results indicate that the resulting method is the current state-of-art LP solver for some classes of problems. Our code is available at https://pub.ist.ac.at/~vnk/papers/IN-FACE-FW.html.

Pavel Arkhipov, Vladimir Kolmogorov

We implement an algorithm for solving the minimum weight perfect matching problem. Our code significantly outperforms the current state-of-the-art Blossom V algorithm on those families of instances where Blossom V takes superlinear time. In practice, our implementation shows almost-linear runtime on every family of instances on which we have tested it. Our algorithm relies on solving the maximum-cardinality unweighted matching problems during its primal phase. Following the state-of-the-art cherry blossom algorithm, we use cherry trees instead of traditional alternating trees and cherry blossoms instead of traditional blossoms. We shrink cherry blossoms rather than traditional blossoms into supernodes. This strategy allows us to deal with much shallower supernodes.

David G. Harris, Vladimir Kolmogorov, Hongyang Liu et al.

The computational equivalence between approximate counting and sampling is well established for polynomial-time algorithms. The most efficient general reduction from counting to sampling is achieved via simulated annealing, where the counting problem is formulated in terms of estimating the ratio $Q={Z(β_{\max})}/{Z(β_{\min})}$ between partition functions $Z(β)=\sum_{x\in Ω} \exp(βH(x))$ of Gibbs distributions $μ_β$ over $Ω$ with Hamiltonian $H$, given access to a sampling oracle that produces samples from $μ_β$ for $β\in [β_{\min}, β_{\max}]$. The best bound achieved by known annealing algorithms with relative error $\varepsilon$ is $O(q \log h / \varepsilon^2)$, where $q, h$ are parameters which respectively bound $\ln Q$ and $H$. However, all known algorithms attaining this near-optimal complexity are inherently sequential, or *adaptive*: the queried parameters $β$ depend on previous samples. We develop a simple non-adaptive algorithm for approximate counting using $O(q \log^2 h / \varepsilon^2)$ samples, as well as an algorithm that achieves $O(q \log h / \varepsilon^2)$ samples with just two rounds of adaptivity, matching the best sample complexity of sequential algorithms. These algorithms naturally give rise to work-efficient parallel (RNC) counting algorithms. We discuss applications to RNC counting algorithms for several classic models, including the anti-ferromagnetic 2-spin, monomer-dimer and ferromagnetic Ising models.

Martin Dvorak, Vladimir Kolmogorov

Farkas established that a system of linear inequalities has a solution if and only if we cannot obtain a contradiction by taking a linear combination of the inequalities. We state and formally prove several Farkas-like theorems over linearly ordered fields in Lean 4. Furthermore, we extend duality theory to the case when some coefficients are allowed to take "infinite values".

Vladimir Kolmogorov, Jack Spalding-Jamieson

We present a family of fast pseudo-approximation algorithms for the minimum balanced vertex separator problem in a graph. Given a graph $G=(V,E)$ with $n$ vertices and $m$ edges, and a (constant) balance parameter $c\in(0,1/2)$, where $G$ has some (unknown) $c$-balanced vertex separator of size ${\rm OPT}_c$, we give a (Monte-Carlo randomized) algorithm running in $O(n^{O(\varepsilon)}m^{1+o(1)})$ time that produces a $Θ(1)$-balanced vertex separator of size $O({\rm OPT}_c\cdot\sqrt{(\log n)/\varepsilon})$ for any value $\varepsilon\in[Θ(1/\log(n)),Θ(1)]$. In particular, for any function $f(n)=ω(1)$ (including $f(n)=\log\log n$, for instance), we can produce a vertex separator of size $O({\rm OPT}_c\cdot\sqrt{\log n}\cdot f(n))$ in time $O(m^{1+o(1)})$. Moreover, for an arbitrarily small constant $\varepsilon=Θ(1)$, our algorithm also achieves the best-known approximation ratio for this problem in $O(m^{1+Θ(\varepsilon)})$ time. The algorithms are based on a semidefinite programming (SDP) relaxation of the problem, which we solve using the Matrix Multiplicative Weight Update (MMWU) framework of Arora and Kale. Our oracle for MMWU uses $O(n^{O(\varepsilon)}\text{polylog}(n))$ almost-linear time maximum-flow computations, and would be sped up if the time complexity of maximum-flow improves.

Paul Swoboda, Vladimir Kolmogorov

We present a new proximal bundle method for Maximum-A-Posteriori (MAP) inference in structured energy minimization problems. The method optimizes a Lagrangean relaxation of the original energy minimization problem using a multi plane block-coordinate Frank-Wolfe method that takes advantage of the specific structure of the Lagrangean decomposition. We show empirically that our method outperforms state-of-the-art Lagrangean decomposition based algorithms on some challenging Markov Random Field, multi-label discrete tomography and graph matching problems.

Pritish Mohapatra, Michal Rolinek, C. V. Jawahar et al.

The accuracy of information retrieval systems is often measured using complex loss functions such as the average precision (AP) or the normalized discounted cumulative gain (NDCG). Given a set of positive and negative samples, the parameters of a retrieval system can be estimated by minimizing these loss functions. However, the non-differentiability and non-decomposability of these loss functions does not allow for simple gradient based optimization algorithms. This issue is generally circumvented by either optimizing a structured hinge-loss upper bound to the loss function or by using asymptotic methods like the direct-loss minimization framework. Yet, the high computational complexity of loss-augmented inference, which is necessary for both the frameworks, prohibits its use in large training data sets. To alleviate this deficiency, we present a novel quicksort flavored algorithm for a large class of non-decomposable loss functions. We provide a complete characterization of the loss functions that are amenable to our algorithm, and show that it includes both AP and NDCG based loss functions. Furthermore, we prove that no comparison based algorithm can improve upon the computational complexity of our approach asymptotically. We demonstrate the effectiveness of our approach in the context of optimizing the structured hinge loss upper bound of AP and NDCG loss for learning models for a variety of vision tasks. We show that our approach provides significantly better results than simpler decomposable loss functions, while requiring a comparable training time.

Vladimir Kolmogorov, Thomas Pock, Michal Rolinek

We consider the problem of minimizing the continuous valued total variation subject to different unary terms on trees and propose fast direct algorithms based on dynamic programming to solve these problems. We treat both the convex and the non-convex case and derive worst case complexities that are equal or better than existing methods. We show applications to total variation based 2D image processing and computer vision problems based on a Lagrangian decomposition approach. The resulting algorithms are very efficient, offer a high degree of parallelism and come along with memory requirements which are only in the order of the number of image pixels.

Neel Shah, Vladimir Kolmogorov, Christoph H. Lampert

Structural support vector machines (SSVMs) are amongst the best performing models for structured computer vision tasks, such as semantic image segmentation or human pose estimation. Training SSVMs, however, is computationally costly, because it requires repeated calls to a structured prediction subroutine (called \emph{max-oracle}), which has to solve an optimization problem itself, e.g. a graph cut. In this work, we introduce a new algorithm for SSVM training that is more efficient than earlier techniques when the max-oracle is computationally expensive, as it is frequently the case in computer vision tasks. The main idea is to (i) combine the recent stochastic Block-Coordinate Frank-Wolfe algorithm with efficient hyperplane caching, and (ii) use an automatic selection rule for deciding whether to call the exact max-oracle or to rely on an approximate one based on the cached hyperplanes. We show experimentally that this strategy leads to faster convergence to the optimum with respect to the number of requires oracle calls, and that this translates into faster convergence with respect to the total runtime when the max-oracle is slow compared to the other steps of the algorithm. A publicly available C++ implementation is provided at http://pub.ist.ac.at/~vnk/papers/SVM.html .

Rustem Takhanov, Vladimir Kolmogorov

We consider two models for the sequence labeling (tagging) problem. The first one is a {\em Pattern-Based Conditional Random Field }(\PB), in which the energy of a string (chain labeling) $x=x_1\ldots x_n\in D^n$ is a sum of terms over intervals $[i,j]$ where each term is non-zero only if the substring $x_i\ldots x_j$ equals a prespecified word $w\in Λ$. The second model is a {\em Weighted Context-Free Grammar }(\WCFG) frequently used for natural language processing. \PB and \WCFG encode local and non-local interactions respectively, and thus can be viewed as complementary. We propose a {\em Grammatical Pattern-Based CRF model }(\GPB) that combines the two in a natural way. We argue that it has certain advantages over existing approaches such as the {\em Hybrid model} of Bened{í} and Sanchez that combines {\em $\mbox{$N$-grams}$} and \WCFGs. The focus of this paper is to analyze the complexity of inference tasks in a \GPB such as computing MAP. We present a polynomial-time algorithm for general \GPBs and a faster version for a special case that we call {\em Interaction Grammars}.

Vladimir Kolmogorov, Christoph Lampert, Emilie Morvant et al.

The 38th Annual Workshop of the Austrian Association for Pattern Recognition (ÖAGM) will be held at IST Austria, on May 22-23, 2014. The workshop provides a platform for researchers and industry to discuss traditional and new areas of computer vision. This year the main topic is: Pattern Recognition: interdisciplinary challenges and opportunities.

Igor Gridchyn, Vladimir Kolmogorov

The problem of minimizing the Potts energy function frequently occurs in computer vision applications. One way to tackle this NP-hard problem was proposed by Kovtun [19,20]. It identifies a part of an optimal solution by running $k$ maxflow computations, where $k$ is the number of labels. The number of "labeled" pixels can be significant in some applications, e.g. 50-93% in our tests for stereo. We show how to reduce the runtime to $O(\log k)$ maxflow computations (or one {\em parametric maxflow} computation). Furthermore, the output of our algorithm allows to speed-up the subsequent alpha expansion for the unlabeled part, or can be used as it is for time-critical applications. To derive our technique, we generalize the algorithm of Felzenszwalb et al. [7] for {\em Tree Metrics}. We also show a connection to {\em $k$-submodular functions} from combinatorial optimization, and discuss {\em $k$-submodular relaxations} for general energy functions.

Vladimir Kolmogorov

We propose a new family of message passing techniques for MAP estimation in graphical models which we call {\em Sequential Reweighted Message Passing} (SRMP). Special cases include well-known techniques such as {\em Min-Sum Diffusion} (MSD) and a faster {\em Sequential Tree-Reweighted Message Passing} (TRW-S). Importantly, our derivation is simpler than the original derivation of TRW-S, and does not involve a decomposition into trees. This allows easy generalizations. We present such a generalization for the case of higher-order graphical models, and test it on several real-world problems with promising results.

Carl Olsson, Johannes Ulen, Yuri Boykov et al.

Energies with high-order non-submodular interactions have been shown to be very useful in vision due to their high modeling power. Optimization of such energies, however, is generally NP-hard. A naive approach that works for small problem instances is exhaustive search, that is, enumeration of all possible labelings of the underlying graph. We propose a general minimization approach for large graphs based on enumeration of labelings of certain small patches. This partial enumeration technique reduces complex high-order energy formulations to pairwise Constraint Satisfaction Problems with unary costs (uCSP), which can be efficiently solved using standard methods like TRW-S. Our approach outperforms a number of existing state-of-the-art algorithms on well known difficult problems (e.g. curvature regularization, stereo, deconvolution); it gives near global minimum and better speed. Our main application of interest is curvature regularization. In the context of segmentation, our partial enumeration technique allows to evaluate curvature directly on small patches using a novel integral geometry approach.

Rustem Takhanov, Vladimir Kolmogorov

We consider Conditional Random Fields (CRFs) with pattern-based potentials defined on a chain. In this model the energy of a string (labeling) $x_1...x_n$ is the sum of terms over intervals $[i,j]$ where each term is non-zero only if the substring $x_i...x_j$ equals a prespecified pattern $α$. Such CRFs can be naturally applied to many sequence tagging problems. We present efficient algorithms for the three standard inference tasks in a CRF, namely computing (i) the partition function, (ii) marginals, and (iii) computing the MAP. Their complexities are respectively $O(n L)$, $O(n L \ell_{max})$ and $O(n L \min\{|D|,\log (\ell_{max}+1)\})$ where $L$ is the combined length of input patterns, $\ell_{max}$ is the maximum length of a pattern, and $D$ is the input alphabet. This improves on the previous algorithms of (Ye et al., 2009) whose complexities are respectively $O(n L |D|)$, $O(n |Γ| L^2 \ell_{max}^2)$ and $O(n L |D|)$, where $|Γ|$ is the number of input patterns. In addition, we give an efficient algorithm for sampling. Finally, we consider the case of non-positive weights. (Komodakis & Paragios, 2009) gave an $O(n L)$ algorithm for computing the MAP. We present a modification that has the same worst-case complexity but can beat it in the best case.

Vladimir Kolmogorov, Martin Wainwright

Tree-reweighted max-product (TRW) message passing is a modified form of the ordinary max-product algorithm for attempting to find minimal energy configurations in Markov random field with cycles. For a TRW fixed point satisfying the strong tree agreement condition, the algorithm outputs a configuration that is provably optimal. In this paper, we focus on the case of binary variables with pairwise couplings, and establish stronger properties of TRW fixed points that satisfy only the milder condition of weak tree agreement (WTA). First, we demonstrate how it is possible to identify part of the optimal solution|i.e., a provably optimal solution for a subset of nodes| without knowing a complete solution. Second, we show that for submodular functions, a WTA fixed point always yields a globally optimal solution. We establish that for binary variables, any WTA fixed point always achieves the global maximum of the linear programming relaxation underlying the TRW method.

Vladimir Kolmogorov, Thomas Schoenemann

This paper addresses the problem of approximate MAP-MRF inference in general graphical models. Following [36], we consider a family of linear programming relaxations of the problem where each relaxation is specified by a set of nested pairs of factors for which the marginalization constraint needs to be enforced. We develop a generalization of the TRW-S algorithm [9] for this problem, where we use a decomposition into junction chains, monotonic w.r.t. some ordering on the nodes. This generalizes the monotonic chains in [9] in a natural way. We also show how to deal with nested factors in an efficient way. Experiments show an improvement over min-sum diffusion, MPLP and subgradient ascent algorithms on a number of computer vision and natural language processing problems.