Approximate inference of marginals using the IBIA framework

Exact inference of marginals in probabilistic graphical models (PGM) is known to be intractable, necessitating the use of approximate methods. Most of the existing variational techniques perform iterative message passing in loopy graphs which is slow to converge for many benchmarks. In this paper, we propose a new algorithm for marginal inference that is based on the incremental build-infer-approximate (IBIA) paradigm. Our algorithm converts the PGM into a sequence of linked clique tree forests (SLCTF) with bounded clique sizes, and then uses a heuristic belief update algorithm to infer the marginals. For the special case of Bayesian networks, we show that if the incremental build step in IBIA uses the topological order of variables then (a) the prior marginals are consistent in all CTFs in the SLCTF and (b) the posterior marginals are consistent once all evidence variables are added to the SLCTF. In our approach, the belief propagation step is non-iterative and the accuracy-complexity trade-off is controlled using user-defined clique size bounds. Results for several benchmark sets from recent UAI competitions show that our method gives either better or comparable accuracy than existing variational and sampling based methods, with smaller runtimes.

Paper

Similar papers

Peer review

Reviewer QKYy4/10 · confidence 5/52023-07-04

Summary

The paper presents a new approximation scheme for computing posterior marginals in graphical models such as Bayesian networks and Markov networks. The proposed approach is based on a previously introduced scheme called IBIA (Incremental Build Infer Approximate) which was initially developed for approximating the partition function. The idea is to construct a sequence of bounded size clique trees (called a clique tree forest) that eventually spans all the input factors and subsequently compute the posterior marginals by carefully selecting the appropriate cliques from the forest. The experimental evaluation is carried out on standard benchmarks for graphical models and the proposed scheme is compared with several state-of-the-art approximation scheme based on message passing. The results demonstrate that the proposed approximation scheme is competitive and some times outperforms existing approaches.

Strengths

The paper is relatively well written and organised. It is also relatively self contained and most of the concepts are discussed in sufficient detail. The empirical evaluation is sound and the results are presented in a relatively clear manner. Some theoretical properties of the proposed method are also discussed.

Weaknesses

Firstly, the contribution presented in this paper appears to be quite incremental over the previous work the IBIA framework. Namely it consists of the message passing scheme over the additional links between adjacent clique trees. The proposed heuristic for choosing links appears to be quite arbitrary. Secondly, I think the proposed method is highly sensitive to the order in which factors are added to the clique tree forest, as well as the initial collection of factors chosen for initialisation. Therefore, the paper should discuss this aspect in more detail. Thirdly, the paper does not mention anything about the convergence of the scheme nor its time/space complexity.

Questions

1. As presented, the performance measure considered assumes that the exact marginals are available. However, many of the problem instances from the experimental section have large induced widths (e.g., 37 for GridBNs) and therefore are difficult to solve exactly, unless a lot of evidence is asserted in the network. Therefore, how many evidence variables are considered for each of the benchmark classes? And also, how do you compute the score when exact inference results are not available? 2. What's the reason for defining the clique size in terms of log instead of just the number of variables in the clique? I found the log based definition confusing. 3. Why do we need to consider two bounds on the clique size? I think the scheme is well defined with only one.

Rating

4: Borderline reject: Technically solid paper where reasons to reject, e.g., limited evaluation, outweigh reasons to accept, e.g., good evaluation. Please use sparingly.

Confidence

5: You are absolutely certain about your assessment. You are very familiar with the related work and checked the math/other details carefully.

Soundness

3 good

Presentation

3 good

Contribution

2 fair

Limitations

The proposed scheme is clearly related to several existing scheme that aim to bound the size of the join-tree. Prominent examples are the mini-bucket scheme for variable elimination, the mini-clustering approach for cluster tree (join tree) elimination, as well as the iterative join-graph propagation scheme (all from Rina Dechter's group). Similarly, the Incremental Thin Junction Tree scheme by Frank Hutter is also looking at bounding the clique size of a join tree by splitting a larger clique into smaller cliques. Therefore, I think a deeper discussion on the connection between these previous schemes and the proposed IBIA based one is clearly called for (it is currently missing from the paper). The empirical evaluation only considers a subset of the existing message passing schemes for approximating marginals e.g, LBP, IJGP, and SampleSearch. These methods are more than one decade old (if not more), do not really fall into the more recent category of variational inference methods. I think it would be interesting to compare with more recent methods based on weighted mini-buckets and cost-shifting such as those developed in Alex Ihler's group (including the importance sampling based ones). If you look at the recent UAI-2022 competition results, it looks like these kinds of methods performed best in the MAP (and to some extent in the PR category as well).

Reviewer SJmy6/10 · confidence 4/52023-07-06

Summary

The paper proposes a new approximate inference algorithm for probabilistic graphical models (PGMs). The approach uses a framework called IBIA to perform marginal inference in discrete PGMs. The main idea is to perform Belief Propagation on a sequence of clique trees. In each step the clique tree is approximated such that the number of variables in any factor is less than a bound and BP is performed on this approximate tree. Factors are then added to this tree and the messages are updated using heuristics to get the next clique tree in the sequence.

Strengths

The approach proposed in the paper seems to work well over a large set of UAI benchmarks in comparison with well-known inference methods. The idea of computing marginals based on a sequence of linked Clique Tree Forests seems novel. The idea of running BP for a single iteration to calibrate and obtaining good experimental results seems interesting since it can reduce computational complexity.

Weaknesses

The presentation could be improved since it seems a bit hard to understand all the steps. Perhaps a pseudo-code like algorithm might have helped. Also, since there have been a lot of approximate inference algorithms in PGMs, the main motivation for this approach was a bit unclear. For example, if I understand correctly the mini-bucket methods (Dechter 2003) also have a similar idea of approximating factors through mini-buckets. I think in the related work section the significance of the method can be perhaps better motivated in relation to these methods. One aspect that was not very clear was when BP converges on the approximate clique tree, if we change the cliques (by adding factors) we cannot just update the beliefs without performing inference again I assumed. In that case, if we use heuristics to update the beliefs (as is mentioned) won’t the errors propagate through the sequence of clique trees? I think there are convergence guarantees on he marginals (e.g. as we add more clique trees in the sequence), that seems like a significant drawback since the scheme will not be any-time (as compared to several other techniques which have this property).

Questions

What is the effect of longer vs shorter sequences, how will this affect the experiment results? How is the idea conceptually related to other methods such as mini-buckets, etc.?

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

2 fair

Contribution

2 fair

Limitations

Limitations are mentioned.

Reviewer vgXS7/10 · confidence 3/52023-07-06

Summary

The paper proposes a new approximate inference method for marginals of probabilistic graphical models (PGMs) based on the incremental build-infer-approximate (IBIA) framework. IBIA was introduced by Bathla & Vasudevan in an arXiv preprint in April 2023 to estimate the partition function (normalizing constant). It achieves this by converting the PGM into a sequence of clique tree forests (SCTFs) with a user-defined clique size bound. The sequence is constructed iteratively: the first step (Build) adds factors of the PGM to the current CTF until the maximum clique size is reached; the second step (Infer) calibrates all clique trees using standard belief propagation; and the final step (Approximate) reduces the clique size to another user-defined bound. This paper builds upon the previous IBIA work in the following ways: it observes and proves several properties of the SCTF generated by the IBIA framework, which it uses to design an extension of IBIA to approximate marginals (in addition to the partition function). This extension works by introducing links between cliques in subsequent steps of the SCTF based on shared variables, called sequence of linked CTFs (SLCTF). It backpropagates beliefs from the last CTF to previous CTFs via the links and re-calibrates each CTF with one round of message passing. Finally the marginal distribution can be inferred from the CTF containing the desired variable. The paper also describes heuristics for the choice of links and notes that in the directed case (i.e. Bayesian networks), it is advantageous to add variables in topological order during the incremental build step. The experimental evaluation compares with two existing methods, loopy belief propagation (LBP) and iterative join graph propagation (IJGP), on benchmarks from recent UAI competitions with time limits of 2 min, 20 min, and, for some benchmarks, 1h. They report the number of instances solved within the time limit and the average and maximum Hellinger distance over all non-evidence variables. The new method performs performs better than LBP and IJGP on most examples and is similar on the remaining ones, even though IBIA is implemented in Python whereas the baselines are written in C++.

Strengths

The proposed approximate inference method is interesting and improves the performance compared to existing methods (LBP and IJGP). The extension of the IBIA framework with links is new and useful. The explanation is adequate given the page constraints and the use of a running example is very helpful to understand the algorithm.

Weaknesses

The contribution of this paper (adding links) seems to me to be relatively minor compared to the introduction of IBIA itself in previous work. Extending it in some way to relate successive CTFs (here done via links) seems quite obvious (although that might be the benefit of hindsight). Regarding the experimental evaluation, I had a look at the recent UAI 2022 competition, since the paper mentions that the same benchmarks were used for recent UAI competitions, if I understand line 14 correctly. It looks like IBIA was a competitor there and while performing well, it did not come in first, as the experimental evaluation of this paper might suggest. For this reason, I'm not convinced that the baselines in the experimental section are representative of the field of PGM inference and whether comparisons with more methods are needed to get an accurate picture of IBIA's performance (see my question to the authors below). As a minor point: while the presentation is generally fine, I found the notation confusing at times, see below.

Questions

(I will update my rating if the author's answers address my concerns.) My main question is about the UAI 2022 competition: it seems to me that other submissions there performed better than the baselines in this paper (LBP, IJGP). Is this understanding correct? If so, what is the reason for choosing what seem to be inferior baselines? I was also confused by line 375, which seems to say that topological ordering gives large errors sometimes. Previously, it sounded like topological ordering was essential for good performance and for the theoretical results in Section 3.1 (line 163). Can you clarify the drawbacks and benefits of topological ordering? Do you know why it causes large errors sometimes? I also noticed some confusing notation, which I mention here for lack of a better place in the review form: - equation (1): (i - j) looks like subtraction. In graph theory, an edge is often written as a pair (i, j) instead. - equation (3): I don't understand the point of the quantifier ($\forall$) in the product notation and have never seen this before. - line 100: $CTF_{k-1,a}$ first looked like a double-index to me, but I assume the "a" just stands for "approximation". Typos: - line 69: I think it should be $Scope(\phi_\alpha)$ - Figure 1, incremental build: I think it should be "while" instead of "until"

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

The authors have adequately discussed the limitations of their work.

Reviewer vgXS2023-08-15

Thank you for providing such an extensive response. My concerns regarding the baselines and the UAI2022 competition have been fully addressed and I don't see a problem with the evaluation anymore. Please add your response to the discussion of the topological ordering in the paper; it clarifies a lot. I have raised my rating from 5 to 7 as a result of the rebuttal.

Authorsrebuttal2023-08-17

Thank you for your prompt response. We will certainly add the discussion on topological ordering to the revised version.

Program Chairsdecision2023-09-21

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC