In Machine Learning, an accepted definition of fairness of a decision taken by a classifier is that it should not depend on protected features, such as gender. Unfortunately, when constraints exist between features, such dependencies can be obscured by the constraints. To avoid this problem, we propose that a decision be considered fair if it has a fair explanation. We define a fair explanation as a prime-implicant reason for the decision that does not contain any protected feature (where the constraints are taken into account in the definition of prime-implicant). Surprisingly, ignoring constraints can completely change the fairness of a decision (according to this definition) even in the absence of constraints between protected and unprotected features. Three possible definitions of fairness of a classifier are that for all its decisions (1) there are only fair explanations, (2) there is at least one fair explanation, or (3) changing protected features does not change the outcome. We identify the relationships between these different definitions of fairness and study the computational complexity of testing fairness of classifiers.

We study multivariate decision trees (MDTs), in particular, classes of MDTs determined by the language of relations that can be used to split feature space. An abductive explanation (AXp) of the classification of a particular instance, viewed as a set of feature-value assignments, is a minimal subset of the instance which is sufficient to lead to the same decision. We investigate when finding a single AXp is tractable. We identify tractable languages for real, integer and boolean features. Indeed, in the case of boolean languages, we provide a P/NP-hard dichotomy. We extend this dichotomy to languages defined by formulas whose literals correspond to splits of ordered domains of arbitrary finite size. Experiments indicate that MDTs can provide more compact models than classical decision trees while conserving accuracy and explainability.

A classifier is considered interpretable if each of its decisions has an explanation which is small enough to be easily understood by a human user. A DNF can be seen as a binary classifier kappa over boolean domains. The size of an explanation of a positive decision taken by a DNF kappa is bounded by the size of the terms in kappa, since we can explain a positive decision by giving a term of kappa that evaluates to true. Since both positive and negative decisions must be explained, we consider that interpretable DNFs are those kappa for which both kappa and its complement can be expressed as DNFs composed of terms of bounded size. In this paper, we investigate the family of k-DNFs whose complements can also be expressed as k-DNFs. We compare two such families, namely depth-k decision trees and nested k-DNFs, a novel family of models. Experimental evidence indicates that nested k-DNFs are an interesting alternative to decision trees in terms of interpretability and accuracy.

Explaining decisions of black-box classifiers is both important and computationally challenging. In this paper, we scrutinize explainers that generate feature-based explanations from samples or datasets. We start by presenting a set of desirable properties that explainers would ideally satisfy, delve into their relationships, and highlight incompatibilities of some of them. We identify the entire family of explainers that satisfy two key properties which are compatible with all the others. Its instances provide sufficient reasons, called weak abductive explanations. We then unravel its various sub-families that satisfy subsets of compatible properties. Indeed, we fully characterize all the explainers that satisfy any subset of compatible properties. In particular, we introduce the first (broad family of) explainers that guarantee the existence of explanations and their global consistency. We discuss some of its instances including the irrefutable explainer and the surrogate explainer whose explanations can be found in polynomial time.

History eXplanation based on Predicates (HXP), studies the behavior of a Reinforcement Learning (RL) agent in a sequence of agent’s interactions with the environment (a history), through the prism of an arbitrary predicate [21]. To this end, an action importance score is computed for each action in the history. The explanation consists in displaying the most important actions to the user. As the calculation of an action’s importance is #W[1]-hard, it is necessary for long histories to approximate the scores, at the expense of their quality. We therefore propose a new HXP method, called Backward-HXP, to provide explanations for these histories without having to approximate scores. Experiments show the ability of B-HXP to summarise long histories.

ABSTRACT Determining whether two STRIPS planning instances are isomorphic is the simplest form of comparison between planning instances. It is also a particular case of the problem concerned with finding an isomorphism between a planning instance and a sub‐instance of another instance . One application of such a mapping is to efficiently produce a compiled form containing all solutions to from a compiled form containing all solutions to . We also introduce the notion of embedding from an instance to another instance , which allows us to deduce that has no solution‐plan if is unsolvable. In this paper, we study the complexity of these problems. We show that the first is GI‐complete and can thus be solved, in theory, in quasi‐polynomial time. While we prove the remaining problems to be NP‐complete, we propose an algorithm to build an isomorphism when possible. We report extensive experimental trials on benchmark problems that demonstrate conclusively that applying constraint propagation in preprocessing can greatly improve the efficiency of a SAT solver.

Abductive explanations (AXp’s) are widely used for understanding decisions of classifiers. Existing definitions are suitable when features are independent. However, we show that ignoring constraints when they exist between features may lead to an explosion in the number of redundant or superfluous AXp’s. We propose three new types of explanations that take into account constraints and that can be generated from the whole feature space or from a sample (such as a dataset). They are based on a key notion of coverage of an explanation, the set of instances it explains. We show that coverage is powerful enough to discard redundant and superfluous AXp’s. For each type, we analyse the complexity of finding an explanation and investigate its formal properties. The final result is a catalogue of different forms of AXp’s with different complexities and different formal guarantees.

best student paper award

We study multivariate decision trees (MDTs), in particular, classes of MDTs determined by the language of relations that can be used to split feature space. An abductive explanation (AXp) of the classification of a particular instance, viewed as a set of feature-value assignments, is a minimal subset of the instance which is sufficient to lead to the same decision. We investigate when finding a single AXp is tractable. We identify tractable languages for real, integer and boolean features. Indeed, in the case of boolean languages, we provide a P/NP-hard dichotomy.

Abstract Given a machine learning (ML) model and a prediction, explanations can be defined as sets of features which are sufficient for the prediction. In some applications, and besides asking for an explanation, it is also critical to understand whether sensitive features can occur in some explanation, or whether a non-interesting feature must occur in all explanations. This paper starts by relating such queries respectively with the problems of relevancy and necessity in logic-based abduction. The paper then proves membership and hardness results for several families of ML classifiers. Afterwards the paper proposes concrete algorithms for two classes of classifiers. The experimental results confirm the scalability of the proposed algorithms.

Compilation into propositional languages finds a growing number of practical uses, including in constraint programming, diagnosis and machine learning (ML), among others. One concrete example is the use of propositional languages as classifiers, and one natural question is how to explain the predictions made. This paper shows that for classifiers represented with some of the best-known propositional languages, different kinds of explanations can be computed in polynomial time. These languages include deterministic decomposable negation normal form (d-DNNF), and so any propositional language that is strictly less succinct than d-DNNF. Furthermore, the paper describes optimizations, specific to Sentential Decision Diagrams (SDDs), which are shown to yield more efficient algorithms in practice.

Determining whether two STRIPS planning instances are isomorphic is the simplest form of comparison between planning instances. It is also a particular case of the problem concerned with finding an isomorphism between a planning instance P and a sub-instance of another instance P'. One application of such an isomorphism is to efficiently produce a compiled form containing all solutions to P from a compiled form containing all solutions to P'. In this paper, we study the complexity of both problems. We show that the former is GI-complete, and can thus be solved, in theory, in quasi-polynomial time. While we prove the latter to be NP-complete, we propose an algorithm to build an isomorphism, when possible. We report extensive experimental trials on benchmark problems which demonstrate conclusively that applying constraint propagation in preprocessing can greatly improve the efficiency of a SAT solver.

Explaining decisions is at the heart of explainable AI. We investigate the computational complexity of providing a formally-correct and minimal explanation of a decision taken by a classifier. In the case of threshold (i.e. score-based) classifiers, we show that a complexity dichotomy follows from the complexity dichotomy for languages of cost functions. In particular, submodular classifiers allow tractable explanation of positive decisions, but not negative decisions (assuming P≠NP). This is an example of the possible asymmetry between the complexity of explaining positive and negative decisions of a particular classifier. Nevertheless, there are large families of classifiers for which explaining both positive and negative decisions is tractable, such as monotone or linear classifiers. We extend tractable cases to constrained classifiers (when there are constraints on the possible input vectors) and to the search for contrastive rather than abductive explanations. Indeed, we show that tractable classes coincide for abductive and contrastive explanations in the constrained or unconstrained settings.

In many classification tasks there is a requirement of monotonicity. Concretely, if all else remains constant, increasing (resp. decreasing) the value of one or more features must not decrease (resp. increase) the value of the prediction. Despite comprehensive efforts on learning monotonic classifiers, dedicated approaches for explaining monotonic classifiers are scarce and classifier-specific. This paper describes novel algorithms for the computation of one formal explanation of a (black-box) monotonic classifier. These novel algorithms are polynomial (indeed linear) in the run time complexity of the classifier. Furthermore, the paper presents a practically efficient model-agnostic algorithm for enumerating formal explanations.

We investigate rules which allow variable elimination in binary CSP (constraint satisfaction problem) instances while conserving satisfiability. We propose new rules and compare them, both theoretically and experimentally. We give optimised algorithms to apply these rules and show that each defines a novel tractable class. Using our variable-elimination rules in preprocessing allowed us to solve more benchmark problems than without.

We study a simple version of multi-agent epistemic planning where the number of parallel steps has to be minimized. We prove that this extension of classical planning is in PSPACE. We propose an encoding in PDDL and present some experiments providing evidence that this encoding allows us to solve practical problems. The types of problems we can encode include problems in which one agent can teach another agent how to perform a task and communication problems where some information must not be revealed to some agents.

Recent work proposed the computation of so-called PI-explanations of Naive Bayes Classifiers (NBCs). PI-explanations are subset-minimal sets of feature-value pairs that are sufficient for the prediction, and have been computed with state-of-the-art exact algorithms that are worst-case exponential in time and space. In contrast, we show that the computation of one PI-explanation for an NBC can be achieved in log-linear time, and that the same result also applies to the more general class of linear classifiers. Furthermore, we show that the enumeration of PI-explanations can be obtained with polynomial delay. Experimental results demonstrate the performance gains of the new algorithms when compared with earlier work. The experimental results also investigate ways to measure the quality of heuristic explanations.

Graphical models (GMs) define a family of mathematical models aimed at the concise description of multivariate functions using decomposability. We restrict ourselves to functions of discrete variables but try to cover a variety of models that are not always considered as "Graphical Models", ranging from functions with Boolean variables and Boolean co-domain (used in automated reasoning) to functions over finite domain variables and integer or real co-domains (usual in machine learning and statistics). We use a simple algebraic semi-ring based framework for generality, define associated queries, relationships between graphical models, complexity results, and families of algorithms, with their associated guarantees.

Singleton arc consistency is an important type of local consistency which has been recently shown to solve all constraint satisfaction problems (CSPs) over constraint languages of bounded width. We aim to characterise all classes of CSPs defined by a forbidden pattern that are solved by singleton arc consistency and closed under removing constraints. We identify five new patterns whose absence ensures solvability by singleton arc consistency, four of which are provably maximal and three of which generalise 2-SAT. Combined with simple counter-examples for other patterns, we make significant progress towards a complete classification.

Singleton arc consistency is an important type of local consistency which has been recently shown to solve all constraint satisfaction problems (CSPs) over constraint languages of bounded width. We aim to characterise all classes of CSPs defined by a forbidden pattern that are solved by singleton arc consistency and closed under removing constraints. We identify five new patterns whose absence ensures solvability by singleton arc consistency, four of which are provably maximal and three of which generalise 2-SAT. Combined with simple counter-examples for other patterns, we make significant progress towards a complete classification.

Characterising tractable fragments of the constraint satisfaction problem (CSP) is an important challenge in theoretical computer science and artificial intelligence. Forbidding patterns (generic sub-instances) provides a means of defining CSP fragments which are neither exclusively language-based nor exclusively structure-based. It is known that the class of binary CSP instances in which the broken-triangle pattern (BTP) does not occur, a class which includes all tree-structured instances, are decided by arc consistency (AC), a ubiquitous reduction operation in constraint solvers. We provide a characterisation of simple partially-ordered forbidden patterns which have this AC-solvability property. It turns out that BTP is just one of five such AC-solvable patterns. The four other patterns allow us to exhibit new tractable classes.

We study methods for transforming valued constraint satisfaction problems (VCSPs) to binary VCSPs. First, we show that the standard dual encoding preserves many aspects of the algebraic properties that capture the computational complexity of VCSPs. Second, we extend the reduction of CSPs to binary CSPs described by Bulín et al. [ Log. Methods Comput. Sci., 11 (2015)] to VCSPs. This reduction establishes that VCSPs over a fixed valued constraint language are polynomial-time equivalent to minimum-cost homomorphism problems over a fixed digraph.

A realistic model of multi-agent planning must allow us to formalize notions which are absent in classical planning, such as communication and knowledge. We investigate multi-agent planning based on a simple logic of knowledge that is grounded on the visibility of propositional variables. Using such a formal logic allows us to prove the existence of a plan given the description of the individual actions. We present an encoding of multi-agent planning problems expressed in this logic into the standard planning language PDDL. The solvability of a planning task is reduced to a model checking problem in a dynamic extension of our logic, proving its complexity. Feeding the resulting problem into a PDDL planner provides a provably correct plan for the original multi-agent planning problem. We apply our method on several examples such as the gossip problem.

The gossip problem, in which information (secrets) must be shared among a certain number of agents using the minimum number of calls, is of interest in the conception of communication networks and protocols. We extend the gossip problem to arbitrary epistemic depths. For example, we may require not only that all agents know all secrets but also that all agents know that all agents know all secrets. We give optimal protocols for the generalised gossip problem, in the case of two-way communications, one-way communications and parallel communication. In the presence of negative goals testing the existence of a successful protocol is NP-complete.

Characterising tractable fragments of the constraint satisfaction problem (CSP) is an important challenge in theoretical computer science and artificial intelligence. Forbidding patterns (generic sub-instances) provides a means of defining CSP fragments which are neither exclusively language-based nor exclusively structure-based. It is known that the class of binary CSP instances in which the broken-triangle pattern (BTP) does not occur, a class which includes all tree-structured instances, are decided by arc consistency (AC), a ubiquitous reduction operation in constraint solvers. We provide a characterisation of simple partially-ordered forbidden patterns which have this AC-solvability property. It turns out that BTP is just one of five such AC-solvable patterns. The four other patterns allow us to exhibit new tractable classes.

Constraint programming is a natural paradigm for many combinatorial optimisation problems. The complexity of constraint satisfaction for various forms of constraints has been widely-studied, both to inform the choice of appropriate algorithms, and to understand better the boundary between polynomial-time complexity and NP-hardness. In constraint programming it is well-known that any constraint satisfaction problem can be converted to an equivalent binary problem using the so-called dual encoding. Using this standard approach any fixed collection of constraints, of arbitrary arity, can be converted to an equivalent set of constraints of arity at most two. Here we show that this transformation, although it changes the domain of the constraints, preserves all the relevant algebraic properties that determine the complexity. Moreover, we show that the dual encoding preserves many of the key algorithmic properties of the original instance. We also show that this remains true for more general valued constraint languages, where constraints may assign different cost values to different assignments. Hence, we obtain a simple proof of the fact that to classify the computational complexity of all valued constraint languages it suffices to classify only binary valued constraint languages.

This paper describes a polynomially-solvable class of temporal planning problems. Polynomiality follows from two assumptions. Firstly, by supposing that each sub-goal fluent can be established by at most one action, we can quickly determine which actions are necessary in any plan. Secondly, the monotonicity of sub-goal fluents allows us to express planning as an instance of STP≠ (Simple Temporal Problem with difference constraints). This class includes temporally-expressive problems requiring the concurrent execution of actions, with potential applications in the chemical, pharmaceutical and construction industries. We also show that any (temporal) planning problem has a monotone relaxation which can lead to the polynomial-time detection of its unsolvability in certain cases. Indeed we show that our relaxation is orthogonal to relaxations based on the ignore-deletes approach used in classical planning since it preserves deletes and can also exploit temporal information.

Relaxation is ubiquitous in the practical resolution of combinatorial problems. If a valid relaxation of an instance has no solution then the original instance has no solution. A tractable relaxation can be built and solved in polynomial time. The most obvious application is the efficient detection of certain unsolvable instances. We review existing relaxation techniques in temporal planning and propose an alternative relaxation inspired by a tractable class of temporal planning problems. Our approach is orthogonal to relaxations based on the ignore-all-deletes approach used in non-temporal planning. We show that our relaxation can even be applied to non-temporal problems, and can also be used to extend a tractable class of temporal planning problems.

To correctly model certain real‐world planning problems, it is essential to take into account time. This is the case for problems requiring the concurrent execution of actions (known as temporally expressive problems). In this paper, we define and study the notion of temporally cyclic problems, that is problems involving sets of cyclically dependent actions. We characterize those temporal planning languages, which can express temporally cyclic problems. We also present a polynomial‐time algorithm, which transforms a temporally cyclic problem into an equivalent acyclic problem. Applying our transformation allows any temporal planner to solve temporally cyclic problems without explicitly managing cyclicity. We first present our results for temporal PDDL (Planning Domain Description Language) 2.1 and then extend them to a language that allows conditions over arbitrary intervals and effects at arbitrary instants.

Discrete optimization problems arise in many different areas and are studied under many different names. In many such problems the quantity to be optimized can be expressed as a sum of functions of a restricted form. Here we present a unifying theory of complexity for problems of this kind. We show that the complexity of a finite-domain discrete optimization problem is determined by certain algebraic properties of the objective function, which we call weighted polymorphisms. We define a Galois connection between sets of rational-valued functions and sets of weighted polymorphisms and show how the closed sets of this Galois connection can be characterized. These results provide a new approach to studying the complexity of discrete optimization. We use this approach to identify certain maximal tractable subproblems of the general problem and hence derive a complete classification of complexity for the Boolean case.

Novel tractable classes of the binary CSP (constraint satisfaction problem) have recently been discovered by studying classes of instances defined by excluding subproblems described by patterns. The complete characterisation of all tractable classes defined by forbidden patterns is a challenging problem. We demonstrate a dichotomy in the case of forbidden patterns consisting of two constraints.

The constraint satisfaction problem (CSP) is a general problem central to computer science and artificial intelligence. Although the CSP is NP-hard in general, considerable effort has been spent on identifying tractable subclasses. The main two approaches consider structural properties (restrictions on the hypergraph of constraint scopes) and relational properties (restrictions on the language of constraint relations). Recently, some authors have considered hybrid properties that restrict the constraint hypergraph and the relations simultaneously. Our key contribution is the novel concept of a CSP pattern and classes of problems defined by forbidden patterns (which can be viewed as forbidding generic sub-problems). We describe the theoretical framework which can be used to reason about classes of problems defined by forbidden patterns. We show that this framework generalises certain known hybrid tractable classes. Although we are not close to obtaining a complete characterisation concerning the tractability of general forbidden patterns, we prove a dichotomy in a special case: classes of problems that arise when we can only forbid binary negative patterns (generic sub-problems in which only disallowed tuples are specified). In this case we show that all (finite sets of) forbidden patterns define either polynomial-time solvable or NP-complete classes of instances.

The minimisation problem of a sum of unary and pairwise functions of discrete variables is a general NP-hard problem with wide applications such as computing MAP configurations in Markov Random Fields (MRF), minimising Gibbs energy, or solving binary Valued Constraint Satisfaction Problems (VCSPs). We study the computational complexity of classes of discrete optimisation problems given by allowing only certain types of costs in every triangle of variable-value assignments to three distinct variables. We show that for several computational problems, the only non- trivial tractable classes are the well known maximum matching problem and the recently discovered joint-winner property. Our results, apart from giving complete classifications in the studied cases, provide guidance in the search for hybrid tractable classes; that is, classes of problems that are not captured by restrictions on the functions (such as submodularity) or the structure of the problem graph (such as bounded treewidth). Furthermore, we introduce a class of problems with convex cardinality functions on cross-free sets of assignments. We prove that while imposing only one of the two conditions renders the problem NP-hard, the conjunction of the two gives rise to a novel tractable class satisfying the cross-free convexity property, which generalises the joint-winner property to problems of unbounded arity.

Cost-optimal planning, in which the aim is to minimise the sum of costs of actions, is a challenging problem due to its computational complexity. A linear program derived from a relaxation which ignores the constraints on the ordering of actions can be used to obtain a lower bound on the cost of a solution-plan. We show that the dual of this linear program provides a transformation of the problem into an equivalent optimal planning problem in which the cost of the goal-achieving action is exactly equal to this lower bound. This transformation is of universal utility since it can be applied as a preprocessing technique and can thus be combined with any of the diverse techniques which have been developed for optimal planning.

An important aspect of any automatic planner is the language in which the user expresses problem instances. A rich language is an advantage for the user, whereas a simple language is an advantage for the programmer who must write a program to solve all planning problems expressible in the language. Considering the temporal planning language PDDL 2.1 as a low-level language, we show how to automatically compile a much richer language into PDDL 2.1. The worst-case complexity of this transformation is quadratic. Our high-level language allows the user to declare time-points and impose simple temporal constraints between them. Conditions and effects can be imposed at time-points, over intervals and over sliding intervals within fixed intervals. Non-instantaneous transitions can also be modelled.

In order to correctly model certain real-world planning problems, it is essential to take into account time. This is the case for problems requiring the concurrent execution of actions (known as temporally-expressive problems). However, we show in this paper that certain existing planners which solve this type of problem are, in fact, incomplete. They cannot guarantee to find a solution to a problem involving sets of cyclically-dependent actions (which we call temporally-cyclic problems). We characterize those temporal planning languages which can express temporally-cyclic problems. We also present a polynomial-time algorithm which transforms a temporally-cyclic problem into an equivalent acyclic problem. Applying our transformation restores the completeness of these temporal planners.

The constraint satisfaction problem (CSP) is a central generic problem in artificial intelligence. Considerable progress has been made in identifying properties which ensure tractability in such problems, such as the property of being tree-structured. In this paper we introduce the broken-triangle property, which allows us to define a hybrid tractable class for this problem which significantly generalizes the class of problems with tree structure. We show that the broken-triangle property is conservative (i.e., it is preserved under domain reduction and hence under arc consistency operations) and that there is a polynomial-time algorithm to determine an ordering of the variables for which the broken-triangle property holds (or to determine that no such ordering exists). We also present a non-conservative extension of the broken-triangle property which is also sufficient to ensure tractability and can be detected in polynomial time.

We present a discrete labeling scheme for line drawings of curved objects which can be seen as an information-rich extension of the classic line-labeling scheme in which lines are classified as convex, concave, occluding or extremal. New labels are introduced to distinguish between curved and planar surface-patches, to identify orthogonal edges and to indicate gradient directions of planar surface-patches.

A statistical analysis of polysemy in sixteen English and French dictionaries has revealed that, in each dictionary, the number of senses per word has a near-exponential distribution. A probabilistic model of historical semantics is presented which explains this distribution. This mathematical model also provides a means of estimating the average number of distinct concepts per word, which was found to be considerably less than the average number of senses listed per word. The grouping of word senses into concepts is based on whether they could inspire the same new senses (by metaphor, metonymy, etc.), that is, their potential future rather than their history.

Many researchers in artificial intelligence are beginning to explore the use of soft constraints to express a set of (possibly conflicting) problem requirements. A soft constraint is a function defined on a collection of variables which associates some measure of desirability with each possible combination of values for those variables. However, the crucial question of the computational complexity of finding the optimal solution to a collection of soft constraints has so far received very little attention. In this paper we identify a class of soft binary constraints for which the problem of finding the optimal solution is tractable. In other words, we show that for any given set of such constraints, there exists a polynomial time algorithm to determine the assignment having the best overall combined measure of desirability. This tractable class includes many commonly-occurring soft constraints, such as 'as near as possible' or 'as soon as possible after', as well as crisp constraints such as 'greater than'. Finally, we show that this tractable class is maximal, in the sense that adding any other form of soft binary constraint which is not in the class gives rise to a class of problems which is NP-hard.

The measurement of information is potentially as important to an information engineer as the measurement of physical quantities is to a civil or mechanical engineer. This article introduces semantic measures, representing the distance between the meanings of two messages. We demonstrate one possible application by giving a small‐scale example in which a semantic measure was used to retrieve information by meaning rather than by word‐occurrence. The distance function between the meanings of two messages can be generalized to cover fuzzy meanings. A possible application, the processing of free responses to opinion polls, is described.