An artificial neuron with binary inputs and a binary output corresponds to a Boolean function. Hence, to explain and verify the behavior of such a neuron (and by extension, a neural network), it suffices to explain and verify its Boolean function. There has been recent interest in representing the Boolean function of such a neuron as an Ordered Binary Decision Diagram (OBDD), which facilitates such analyses. In this paper, we propose an anytime algorithm for compiling a binary neuron into an OBDD, based on a recently proposed compiler that decomposes a binary neuron's Boolean function into its prime implicants, represented as a decision tree. We augment this compiler so that it outputs an OBDD instead. Our augmented compiler is also anytime, as it produces intermediate OBDDs that represent inner- and outer-bounds of the original neuron, which tighten as compilation progresses. Theoretically, decision graphs of binary neurons are exponentially more succinct than their decision trees. Empirically, compilation to decision graphs can scale to neurons with over a thousand features, compared to dozens of features using other compilers. We highlight the utility of our approach via a case study in eXplainable AI.

Knowledge compilation offers a formal approach to explaining and verifying the behavior of machine learning systems, such as neural networks. Unfortunately, compiling even an individual neuron into a tractable representation such as an Ordered Binary Decision Diagram (OBDD), is an NP-hard problem. In this paper, we consider the problem of training a neuron from data, subject to the constraint that it has a compact representation as an OBDD. Our approach is based on the observation that a neuron can be compiled into an OBDD in polytime if (1) the neuron has integer weights, and (2) its aggregate weight is bounded. Unfortunately, we first show that it is also NP-hard to train a neuron, subject to these two constraints. On the other hand, we show that if we train a neuron generatively, rather than discriminatively, a neuron with bounded aggregate weight can be trained in pseudo-polynomial time. Hence, we propose the first efficient algorithm for training a neuron that is guaranteed to have a compact representation as an OBDD. Empirically, we show that our approach can train neurons with higher accuracy and more compact OBDDs.

We consider the compilation of a binary neural network’s decision function into tractable representations such as Ordered Binary Decision Diagrams (OBDDs) and Sentential Decision Diagrams (SDDs). Obtaining this function as an OBDD/SDD facilitates the explanation and formal verification of a neural network’s behavior. First, we consider the task of verifying the robustness of a neural network, and show how we can compute the expected robustness of a neural network, given an OBDD/SDD representation of it. Next, we consider a more efficient approach for compiling neural networks, based on a pseudo-polynomial time algorithm for compiling a neuron. We then provide a case study in a handwritten digits dataset, highlighting how two neural networks trained from the same dataset can have very high accuracies, yet have very different levels of robustness. Finally, in experiments, we show that it is feasible to obtain compact representations of neural networks as SDDs.

Structured Bayesian networks (SBNs) are a recently proposed class of probabilistic graphical models which integrate background knowledge in two forms: conditional independence constraints and Boolean domain constraints. In this paper, we propose the first exact inference algorithm for SBNs, based on compiling a given SBN to a Probabilistic Sentential Decision Diagram (PSDD). We further identify a tractable subclass of SBNs, which have PSDDs of polynomial size. These SBNs yield a tractable model of route distributions, whose structure can be learned from GPS data, using a simple algorithm that we propose. Empirically, we demonstrate the utility of our inference algorithm, showing that it can be an order-ofmagnitude more efficient than more traditional approaches to exact inference. We demonstrate the utility of our learning algorithm, showing that it can learn more accurate models and classifiers from GPS data.

We propose an algorithm for compiling Bayesian network classifiers into decision graphs that mimic the input and output behavior of the classifiers. In particular, we compile Bayesian network classifiers into ordered decision graphs, which are tractable and can be exponentially smaller in size than decision trees. This tractability facilitates reasoning about the behavior of Bayesian network classifiers, including the explanation of decisions they make. Our compilation algorithm comes with guarantees on the time of compilation and the size of compiled decision graphs. We apply our compilation algorithm to classifiers from the literature and discuss some case studies in which we show how to automatically explain their decisions and verify properties of their behavior.

Testing Bayesian Networks (TBNs) were introduced recently to represent a set of distributions, one of which is selected based on the given evidence and used for reasoning. TBNs are more expressive than classical Bayesian Networks (BNs): Marginal queries correspond to multi-linear functions in BNs and to piecewise multi-linear functions in TBNs. Moreover, TBN queries are universal approximators, like neural networks. In this paper, we study conditional independence in TBNs, showing that it can be inferred from d-separation as in BNs. We also study the role of TBN expressiveness and independence in dealing with the problem of learning with incomplete models (i.e., ones that miss nodes or edges from the data-generating model). Finally, we illustrate our results on a number of concrete examples, including a case study on Hidden Markov Models.

Probabilistic Sentential Decision Diagrams (PSDDs) have been proposed for learning tractable probability distributions from a combination of data and background knowledge (in the form of Boolean constraints). In this paper, we propose a variant on PSDDs, called conditional PSDDs, for representing a family of distributions that are conditioned on the same set of variables. Conditional PSDDs can also be learned from a combination of data and (modular) background knowledge. We use conditional PSDDs to define a more structured version of Bayesian networks, in which nodes can have an exponential number of states, hence expanding the scope of domains where Bayesian networks can be applied. Compared to classical PSDDs, the new representation exploits the independencies captured by a Bayesian network to decompose the learning process into localized learning tasks, which enables the learning of better models while using less computation. We illustrate the promise of conditional PSDDs and structured Bayesian networks empirically, and by providing a case study to the modeling of distributions over routes on a map.

We propose an approach for explaining Bayesian network classifiers, which is based on compiling such classifiers into decision functions that have a tractable and symbolic form. We introduce two types of explanations for why a classifier may have classified an instance positively or negatively and suggest algorithms for computing these explanations. The first type of explanation identifies a minimal set of the currently active features that is responsible for the current classification, while the second type of explanation identifies a minimal set of features whose current state (active or not) is sufficient for the classification. We consider in particular the compilation of Naive and Latent-Tree Bayesian network classifiers into Ordered Decision Diagrams (ODDs), providing a context for evaluating our proposal using case studies and experiments based on classifiers from the literature.

The past decade has seen a significant interest in learning tractable probabilistic representations. Arithmetic circuits (ACs) were among the first proposed tractable representations, with some subsequent representations being instances of ACs with weaker or stronger properties. In this paper, we provide a formal basis under which variants on ACs can be compared, and where the precise roles and semantics of their various properties can be made more transparent. This allows us to place some recent developments on ACs in a clearer perspective and to also derive new results for ACs. This includes an exponential separation between ACs with and without determinism; completeness and incompleteness results; and tractability results (or lack thereof) when computing most probable explanations (MPEs).

Recently, the Probabilistic Sentential Decision Diagram (PSDD) has been proposed as a framework for systematically inducing and learning distributions over structured objects, including combinatorial objects such as permutations and rankings, paths and matchings on a graph, etc. In this paper, we study the scalability of such models in the context of representing and learning distributions over routes on a map. In particular, we introduce the notion of a hierarchical route distribution and show how they can be leveraged to construct tractable PSDDs over route distributions, allowing them to scale to larger maps. We illustrate the utility of our model empirically, in a route prediction task, showing how accuracy can be increased significantly compared to Markov models.

We propose the structured naive Bayes (SNB) classifier, which augments the ubiquitous naive Bayes classifier with structured features. SNB classifiers facilitate the use of complex features, such as combinatorial objects (e.g., graphs, paths and orders) in a general but systematic way. Underlying the SNB classifier is the recently proposed Probabilistic Sentential Decision Diagram (PSDD), which is a tractable representation of probability distributions over structured spaces. We illustrate the utility and generality of the SNB classifier via case studies. First, we show how we can distinguish players of simple games in terms of play style and skill level based purely on observing the games they play. Second, we show how we can detect anomalous paths taken on graphs based purely on observing the paths themselves.

We consider the problem of learning Bayesian network structures from complete data. In particular, we consider the enumeration of their k-best equivalence classes. We propose a new search space for A* search, called the EC graph, that facilitates the enumeration of equivalence classes, by representing the space of completed, partially directed acyclic graphs. We also propose a canonization of this search space, called the EC tree, which further improves the efficiency of enumeration. Empirically, our approach is orders of magnitude more efficient than the state-of-the-art at enumerating equivalence classes.

We consider the problem of learning Bayesian networks optimally, when subject to background knowledge in the form of ancestral constraints. Our approach is based on a recently proposed framework for optimal structure learning based on non-decomposable scores, which is general enough to accommodate ancestral constraints. The proposed framework exploits oracles for learning structures using decomposable scores, which cannot accommodate ancestral constraints since they are non-decomposable. We show how to empower these oracles by passing them decomposable constraints that they can handle, which are inferred from ancestral constraints that they cannot handle. Empirically, we demonstrate that our approach can be orders-of-magnitude more efficient than alternative frameworks, such as those based on integer linear programming.

We consider tractable representations of probability distributions and the polytime operations they support. In particular, we consider a recently proposed arithmetic circuit representation, the Probabilistic Sentential Decision Diagram (PSDD). We show that PSDD supports a polytime multiplication operator, while they do not support a polytime operator for summing-out variables. A polytime multiplication operator make PSDDs suitable for a broader class of applications compared to arithmetic circuits, which do not in general support multiplication. As one example, we show that PSDD multiplication leads to a very simple but effective compilation algorithm for probabilistic graphical models: represent each model factor as a PSDD, and then multiply them.

There are many criteria for measuring the value of information (VOI), each based on a different principle that is usually suitable for specific applications. We propose a new criterion for measuring the value of information, which values information that leads to robust decisions (i.e., ones that are unlikely to change due to new information). We also introduce an algorithm for Naive Bayes networks that selects features with maximal VOI under the new criteria. We discuss the application of the new criteria to classification tasks, showing how it can be used to tradeoff the budget, allotted for acquiring information, with the classification accuracy. In particular, we show empirically that the new criteria can reduce the expended budget significantly while reducing the classification accuracy only slightly. We also show empirically that the new criterion leads to decisions that are much more robust than those based on traditional VOI criteria, such as information gain and classification loss. This make the new criteria particularly suitable for certain decision making applications.

Tractable learning aims to learn probabilistic models where inference is guaranteed to be efficient. However, the particular class of queries that is tractable depends on the model and underlying representation. Usually this class is MPE or conditional probabilities $\Pr(\xs|\ys)$ for joint assignments~$\xs,\ys$. We propose a tractable learner that guarantees efficient inference for a broader class of queries. It simultaneously learns a Markov network and its tractable circuit representation, in order to guarantee and measure tractability. Our approach differs from earlier work by using Sentential Decision Diagrams (SDD) as the tractable language instead of Arithmetic Circuits (AC). SDDs have desirable properties, which more general representations such as ACs lack, that enable basic primitives for Boolean circuit compilation. This allows us to support a broader class of complex probability queries, including counting, threshold, and parity, in polytime.

When making decisions under uncertainty, the optimal choices are often difficult to discern, especially if not enough information has been gathered. Two key questions in this regard relate to whether one should stop the information gathering process and commit to a decision (stopping criterion), and if not, what information to gather next (selection criterion). In this paper, we show that the recently introduced notion, Same-Decision Probability (SDP), can be useful as both a stopping and a selection criterion, as it can provide additional insight and allow for robust decision making in a variety of scenarios. This query has been shown to be highly intractable, being PP^PP-complete, and is exemplary of a class of queries which correspond to the computation of certain expectations. We propose the first exact algorithm for computing the SDP, and demonstrate its effectiveness on several real and synthetic networks. Finally, we present new complexity results, such as the complexity of computing the SDP on models with a Naive Bayes structure. Additionally, we prove that computing the non-myopic value of information is complete for the same complexity class as computing the SDP.

The Sentential Decision Diagram (SDD) is a recently proposed representation of Boolean functions, containing Ordered Binary Decision Diagrams (OBDDs) as a distinguished subclass. While OBDDs are characterized by total variable orders, SDDs are characterized more generally by vtrees. As both OBDDs and SDDs have canonical representations, searching for OBDDs and SDDs of minimal size simplifies to searching for variable orders and vtrees, respectively. For OBDDs, there are effective heuristics for dynamic reordering, based on locally swapping variables. In this paper, we propose an analogous approach for SDDs which navigates the space of vtrees via two operations: one based on tree rotations and a second based on swapping children in a vtree. We propose a particular heuristic for dynamically searching the space of vtrees, showing that it can find SDDs that are an order-of-magnitude more succinct than OBDDs found by dynamic reordering.

EDML is a recently proposed algorithm for learning parameters in Bayesian networks. It was originally derived in terms of approximate inference on a meta-network, which underlies the Bayesian approach to parameter estimation. While this initial derivation helped discover EDML in the first place and provided a concrete context for identifying some of its properties (e.g., in contrast to EM), the formal setting was somewhat tedious in the number of concepts it drew on. In this paper, we propose a greatly simplified perspective on EDML, which casts it as a general approach to continuous optimization. The new perspective has several advantages. First, it makes immediate some results that were non-trivial to prove initially. Second, it facilitates the design of EDML algorithms for new graphical models, leading to a new algorithm for learning parameters in Markov networks. We derive this algorithm in this paper, and show, empirically, that it can sometimes learn better estimates from complete data, several times faster than commonly used optimization methods, such as conjugate gradient and L-BFGS.

The Sentential Decision Diagram (SDD) is a recently proposed representation of Boolean functions, containing Ordered Binary Decision Diagrams (OBDDs) as a distinguished subclass. While OBDDs are characterized by total variable orders, SDDs are characterized by dissections of variable orders, known as vtrees. Despite this generality, SDDs retain a number of properties, such as canonicity and a polytime apply operator, that have been critical to the practical success of OBDDs. Moreover, upper bounds on the size of SDDs were also given, which are tighter than comparable upper bounds on the size of OBDDs. In this paper, we analyze more closely some of the theoretical properties of SDDs and their size. In particular, we consider the impact of basing decisions on sentences (using dissections as in SDDs), in comparison to basing decisions on variables (using total variable orders as in OBDDs). Here, we identify a class of Boolean functions where basing decisions on sentences using dissections of a variable order can lead to exponentially more compact SDDs, compared to OBDDs based on the same variable order. Moreover, we identify a fundamental property of the decompositions that underlie SDDs and use it to show how certain changes to a vtree can also lead to exponential differences in the size of an SDD.

MOTIVATION: Haplotype inference is an important step for many types of analyses of genetic variation in the human genome. Traditional approaches for obtaining haplotypes involve collecting genotype information from a population of individuals and then applying a haplotype inference algorithm. The development of high-throughput sequencing technologies allows for an alternative strategy to obtain haplotypes by combining sequence fragments. The problem of 'haplotype assembly' is the problem of assembling the two haplotypes for a chromosome given the collection of such fragments, or reads, and their locations in the haplotypes, which are pre-determined by mapping the reads to a reference genome. Errors in reads significantly increase the difficulty of the problem and it has been shown that the problem is NP-hard even for reads of length 2. Existing greedy and stochastic algorithms are not guaranteed to find the optimal solutions for the haplotype assembly problem. RESULTS: In this article, we proposed a dynamic programming algorithm that is able to assemble the haplotypes optimally with time complexity O(m x 2(k) x n), where m is the number of reads, k is the length of the longest read and n is the total number of SNPs in the haplotypes. We also reduce the haplotype assembly problem into the maximum satisfiability problem that can often be solved optimally even when k is large. Taking advantage of the efficiency of our algorithm, we perform simulation experiments demonstrating that the assembly of haplotypes using reads of length typical of the current sequencing technologies is not practical. However, we demonstrate that the combination of this approach and the traditional haplotype phasing approaches allow us to practically construct haplotypes containing both common and rare variants.

We introduce a new perspective on approximations to the maximum a posteriori (MAP) task in probabilistic graphical models, that is based on simplifying a given instance, and then tightening the approximation. First, we start with a structural relaxation of the original model. We then infer from the relaxation its deficiencies, and compensate for them. This perspective allows us to identify two distinct classes of approximations. First, we find that max-product belief propagation can be viewed as a way to compensate for a relaxation, based on a particular idealized case for exactness. We identify a second approach to compensation that is based on a more refined idealized case, resulting in a new approximation with distinct properties. We go on to propose a new class of algorithms that, starting with a relaxation, iteratively yields tighter approximations.