Inferring M-Best Diverse Labelings in a Single One

Kirillov, Alexander; Savchynskyy, Bogdan; Schlesinger, Dmitrij; Vetrov, Dmitry; Rother, Carsten

doi:10.1109/iccv.2015.211

Cited by 26 publications

(24 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As a result, the λ m multipliers cannot provide a good trade-off between energy and diversity. Secondly, the incremental (greedy) update of x m does not guarantee that all the m modes are placed in local maxima [11]. Recent joint m-Best methods [12], which do not enforce the MAP solution as part of the set of solutions, outperform sequential ones both in terms of quality and runtime, but the difference in runtime grows with m. Thus, a trade-off between sequential and joint m-best can be implemented by expanding a tree-like structure, such as a binary tree partitioning (BTP) [24] [23], but driven by heuristic variable selection.…”

Section: Sequential Vs Joint M-bestmentioning

confidence: 99%

Seeking affinity structure: Strategies for improving m-best graph matching

Curado

Escolano

Lozano

et al. 2020

Information Sciences

View full text Add to dashboard Cite

State-of-the-art methods for finding the m-best solutions to graph matching (QAP) rely on exclusion strategies. The k-th best solution is found by excluding all better ones from the search space. This provides diversity, a natural requirement for transforming a MAP problem into a m-Best one. Since diversity enforces mode hopping, it is usually combined with a mode-approximation strategy such as marginalisation. However, these methods are generic insofar they do not incorporate the detailed structure of the problem at hand, i.e. the properties of the global affinity matrix which characterise the search space. Without this knowledge, it is thus hard to devise a practical criterion for choosing the next variable to clamp. In this paper, we propose several strategies to select the next variable to clamp, spanning the whole range between depth-first and breadth-first search, and we contribute with a unifying view for characterising the search space on the fly. Our strategies are: a) number of factors in which the variables participate, b) centrality measures associated with the affinity matrix, and c) discrete pooling. Our experiments show that max number of factors and centrality provide a trade-off between efficiency and accuracy, whereas discrete pooling leads to an improvement of the state-of-the-art.

show abstract

Section: Sequential Vs Joint M-bestmentioning

confidence: 99%

Seeking affinity structure: Strategies for improving m-best graph matching

Curado

Escolano

Lozano

et al. 2020

Information Sciences

View full text Add to dashboard Cite

show abstract

“…We use the minimum structural Hamming distance (SHD) between a set of candidate Bayesian networks and the ground truth network to measure the collective capability of the candidate set in discovering the true underlying structure. This is called oracle accuracy, i.e., the best one among the top results, which is commonly used in the literature on multiple diverse predictions (Batra et al 2012;Kirillov et al 2015;Chen et al 2013).…”

Section: Accuracy In Structure Learningmentioning

confidence: 99%

“…Such a diverse set of mode models are expected to provide a much better coverage of true underlying model. This work is inspired by recent success in developing methods for finding multiple diverse predictions, including Diverse M-Best (Batra et al 2012), joint Diverse M-Best (Kirillov et al 2015), and M-Modes (Chen et al 2013;2018). Based on a global-local theorem showing that a mode Bayesian network must be optimal in all local scopes, we introduce an A* search algorithm to efficiently find the top M Bayesian networks which are highly probable and naturally diverse.…”

Section: Introductionmentioning

confidence: 99%

Learning Diverse Bayesian Networks

Chen

Yuan

2019

AAAI

View full text Add to dashboard Cite

Much effort has been directed at developing algorithms for learning optimal Bayesian network structures from data. When given limited or noisy data, however, the optimal Bayesian network often fails to capture the true underlying network structure. One can potentially address the problem by finding multiple most likely Bayesian networks (K-Best) in the hope that one of them recovers the true model. However, it is often the case that some of the best models come from the same peak(s) and are very similar to each other; so they tend to fail together. Moreover, many of these models are not even optimal respective to any causal ordering, thus unlikely to be useful. This paper proposes a novel method for finding a set of diverse top Bayesian networks, called modes, such that each network is guaranteed to be optimal in a local neighborhood. Such mode networks are expected to provide a much better coverage of the true model. Based on a globallocal theorem showing that a mode Bayesian network must be optimal in all local scopes, we introduce an A* search algorithm to efficiently find top M Bayesian networks which are highly probable and naturally diverse. Empirical evaluations show that our top mode models have much better diversity as well as accuracy in discovering true underlying models than those found by K-Best.

show abstract

“…The proposed solution is to sequentially find several local optima and force them to be different from each other by introducing diversity constraints in the objective function. It has recently been shown that it is provably more effective to solve for diverse MAPs jointly but under the same set of constraints [20]. However, none of these methods provide a generic and practical way to choose local constraints to be enforced over variable sub-groups.…”

Section: Mean Field Inferencementioning

confidence: 99%

Multi-modal Mean-Fields via Cardinality-Based Clamping

Baqué

Fleuret

Fua

2017

2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

Mean Field inference is central to statistical physics. It has attracted much interest in the Computer Vision community to efficiently solve problems expressible in terms of large Conditional Random Fields. However, since it models the posterior probability distribution as a product of marginal probabilities, it may fail to properly account for important dependencies between variables.We therefore replace the fully factorized distribution of Mean Field by a weighted mixture of such distributions, that similarly minimizes the KL-Divergence to the true posterior. By introducing two new ideas, namely, conditioning on groups of variables instead of single ones and using a parameter of the conditional random field potentials, that we identify to the temperature in the sense of statistical physics to select such groups, we can perform this minimization efficiently. Our extension of the clamping method proposed in previous works allows us to both produce a more descriptive approximation of the true posterior and, inspired by the diverse MAP paradigms, fit a mixture of Mean Field approximations. We demonstrate that this positively impacts real-world algorithms that initially relied on mean fields.

show abstract

Inferring M-Best Diverse Labelings in a Single One

Cited by 26 publications

References 18 publications

Seeking affinity structure: Strategies for improving m-best graph matching

Seeking affinity structure: Strategies for improving m-best graph matching

Learning Diverse Bayesian Networks

Multi-modal Mean-Fields via Cardinality-Based Clamping

Contact Info

Product

Resources

About