Search

The 2018 AAAI Conference on Artificial Intelligence, the 2018 Innovative Applications of Artificial Intelligence, and the 2018 Symposium on Educational Advances in Artificial Intelligence were held February 2–7, 2018 at the Hilton New Orleans Riverside, New Orleans, Louisiana, USA. This report, based on the prefaces contained in the AAAI-18 proceedings and program, summarizes the events of the conference.

Reinforcement learning (RL) methods usually treat reward functions as black boxes. As such, these methods must extensively interact with the environment in order to discover rewards and optimal policies. In most RL applications, however, users have to program the reward function and, hence, there is the opportunity to make the reward function visible – to show the reward function’s code to the RL agent so it can exploit the function’s internal structure to learn optimal policies in a more sample efficient manner. In this paper, we show how to accomplish this idea in two steps. First, we propose reward machines, a type of finite state machine that supports the specification of reward functions while exposing reward function structure. We then describe different methodologies to exploit this structure to support learning, including automated reward shaping, task decomposition, and counterfactual reasoning with off-policy learning. Experiments on tabular and continuous domains, across different tasks and RL agents, show the benefits of exploiting reward structure with respect to sample efficiency and the quality of resultant policies. Finally, by virtue of being a form of finite state machine, reward machines have the expressive power of a regular language and as such support loops, sequences and conditionals, as well as the expression of temporally extended properties typical of linear temporal logic and non-Markovian reward specification.

Binarized Neural Networks (BNNs) are an important class of neural network characterized by weights and activations restricted to the set \(\{-1,+1\}\). BNNs provide simple compact descriptions and as such have a wide range of applications in low-power devices. In this paper, we investigate a model-based approach to training BNNs using constraint programming (CP), mixed-integer programming (MIP), and CP/MIP hybrids. We formulate the training problem as finding a set of weights that correctly classify the training set instances while optimizing objective functions that have been proposed in the literature as proxies for generalizability. Our experimental results on the MNIST digit recognition dataset suggest that—when training data is limited—the BNNs found by our hybrid approach generalize better than those obtained from a state-of-the-art gradient descent method. More broadly, this work enables the analysis of neural network performance based on the availability of optimal solutions and optimality bounds.

The AAAI-14 Workshop program was held Sunday and Monday, July 27–28, 2012, at the Québec City Convention Centre in Québec, Canada. Canada. The AAAI-14 workshop program included fifteen workshops covering a wide range of topics in artificial intelligence. The titles of the workshops were AI and Robotics; Artificial Intelligence Applied to Assistive Technologies and Smart Environments; Cognitive Computing for Augmented Human Intelligence; Computer Poker and Imperfect Information; Discovery Informatics; Incentives and Trust in Electronic Communities; Intelligent Cinematography and Editing; Machine Learning for Interactive Systems: Bridging the Gap between Perception, Action and Communication; Modern Artificial Intelligence for Health Analytics; Multiagent Interaction without Prior Coordination; Multidisciplinary Workshop on Advances in Preference Handling; Semantic Cities — Beyond Open Data to Models, Standards and Reasoning; Sequential Decision Making with Big Data; Statistical Relational AI; and The World Wide Web and Public Health Intelligence. This article presents short summaries of those events.

This special issue is dedicated to John McCarthy, founding father of Artificial Intelligence. It contains a collection of recent contributions to the field of knowledge representation and reasoning, a field that McCarthy founded and that has been a main focus of his research during the last half century. In this introductory article, we survey some of McCarthy's major contributions to the field of knowledge representation and reasoning, and situate the papers in this special issue in the context of McCarthy's previous work.

In this paper we show how tree decomposition can be applied to reasoning with first-order and propositional logic theories. Our motivation is two-fold. First, we are concerned with how to reason effectively with multiple knowledge bases that have overlap in content. Second, we are concerned with improving the efficiency of reasoning over a set of logical axioms by partitioning the set with respect to some detectable structure, and reasoning over individual partitions either locally or in a distributed fashion. To this end, we provide algorithms for partitioning and reasoning with related logical axioms in propositional and first-order logic.Many of the reasoning algorithms we present are based on the idea of passing messages between partitions. We present algorithms for both forward (data-driven) and backward (query-driven) message passing. Different partitions may have different associated reasoning procedures. We characterize a class of reasoning procedures that ensures completeness and soundness of our message-passing algorithms. We further provide a specialized algorithm for propositional satisfiability checking with partitions. Craig's interpolation theorem serves as a key to proving soundness and completeness of all of these algorithms. An analysis of these algorithms emphasizes parameters of the partitionings that influence the efficiency of computation. We provide a greedy algorithm that automatically decomposes a set of logical axioms into partitions, following this analysis.

Reformulation is ubiquitous in problem solving and is especially common in modeling physical systems. In this paper we examine reformulation techniques in the context of reasoning about physical systems. This paper does not present a general theory of reformulation, but it studies a number of known reformulation techniques to achieve a broad understanding of the space of available reformulations. In doing so, we present a practical framework for specifying, classifying, and evaluating various reformulation techniques applicable to this class of problems. Our framework provides the terminology to specify the conditions under which a particular reformulation technique is applicable, the cost associated with performing the reformulation, and the effects of the reformulation with respect to the problem encoding.

A key element to realizing the Semantic Web is developing a suitably rich language for encoding and describing Web content. Such a language must have a well defined semantics, be sufficiently expressive to describe the complex interrelationships and constraints between Web objects, and be amenable to automated manipulation and reasoning with acceptable limits on time and resource requirements. A key component of the Semantic Web services vision is the creation of a language for describing Web services. DAML-S is such a language it is a DAML+OIL ontology for describing Web services that a coalition of researchers created with support from DARPA.

The authors propose the markup of Web services in the DAML family of Semantic Web markup languages. This markup enables a wide variety of agent technologies for automated Web service discovery, execution, composition and interoperation. The authors present one such technology for automated Web service composition.

This paper provides further contributions to the formal characterization of diagnosis introduced by Reiter and extended by de Kleer, Mackworth and Reiter. An objective of this line of research is to provide a succinct characterization of the potentially exponential space of diagnoses, be it consistency-based or abductive. We have seen that minimal diagnoses provide a parsimonious characterization of the space of consistency-based diagnoses when certain syntactic restrictions are applied to the system description. Otherwise, the space of consistency-based diagnoses must be characterized by the less succinct kernel diagnoses. In this paper we distinguish between minimalfault diagnoses and minimalexonerating diagnoses and show that minimal exonerating diagnoses characterize the space of consistency-based diagnoses under syntactic restrictions related to the use of fault theories. With respect to abduction, we provide a new characterization of the space of diagnoses, distinguishing abductive explanations from abductive diagnoses. The impact of various syntactic restrictions on the characterization of the space of abductive explanations and diagnoses is also examined. By characterizing abductive diagnosis within the same logical framework as consistency-based diagnosis, we are able to employ abduction as a preference criterion to distinguish a subset of consistency-based diagnoses which explain the observations in addition to being consistent with them. As with previous work, we exploit the notions of prime implicants/implicates to provide a link between our formal characterization of diagnostic reasoning and implementation.

In this paper, we address the problem of specifying and computing preferred plans using rich, qualitative, user preferences. We propose a logical language for specifying preferences over the evolution of states and actions associated with a plan. We provide a semantics for our first-order preference language in the situation calculus, and prove that progression of our preference formulae preserves this semantics. This leads to the development of PPlan, a bounded best-first search planner that computes preferred plans. Our preference language is amenable to integration with many existing planners, and beyond planning, can be used to support a diversity of dynamical reasoning tasks that employ preferences.

Automated Planning is an old area of AI that focuses on the development of techniques for finding a plan that achieves a given goal from a given set of initial states as quickly as possible. In most real-world applications, users of planning systems have preferences over the multitude of plans that achieve a given goal. These preferences allow to distinguish plans that are more desirable from those that are less desirable. Planning systems should therefore be able to construct high-quality plans, or at the very least they should be able to build plans that have a reasonably good quality given the resources available.In the last few years we have seen a significant amount of research that has focused on developing rich and compelling languages for expressing preferences over plans. On the other hand, we have seen the development of planning techniques that aim at finding high-quality plans quickly, exploiting some of the ideas developed for classical planning. In this paper we review the latest developments in automated preference-based planning. We also review various approaches for preference representation, and the main practical approaches developed so far.

Mechanisms are decision functions that map the individual preference orderings of separate parties into a single ordering over the group outcome. Unfortunately, classical impossibility results, readily extended to preferences, show that no mechanism can be "fair" for all scenarios [1]. Further, any positive results typically assume that agents do not form coalitions or other such partnerships. While coalitions can complicate both theoretical analysis and underlying paradigms of rationality, in a particular setting they can serve to constrain a problem to the point of circumventing traditional impossibility results. Automated mechanism design (AMD) [2] attempts to overcome such results by designing specific mechanisms for specific situations on the spot. No perfect mechanism exists that works in every context, but whenever there is information about the players, a fair mechanism can exist for that specific setting.

In this paper we consider three different kinds of domain-dependent control knowledge (temporal, procedural and HTN-based) that are useful in planning. Our approach is declarative and relies on the language of logic programming with answer set semantics (AnsProlog*). AnsProlog* is designed to plan without control knowledge. We show how temporal, procedural and HTN-based control knowledge can be incorporated into AnsProlog* by the modular addition of a small number of domain-dependent rules, without the need to modify the planner. We formally prove the correctness of our planner, both in the absence and presence of the control knowledge. Finally, we perform some initial experimentation that demonstrates the potential reduction in planning time that can be achieved when procedural domain knowledge is used to solve planning problems with large plan length., Comment: 70 pages, accepted for publication, TOCL Version with all proofs

In this paper we present a declarative approach to adding domain-dependent control knowledge for Answer Set Planning (ASP). Our approach allows different types of domain-dependent control knowledge such as hierarchical, temporal, or procedural knowledge to be represented and exploited in parallel, thus combining the ideas of control knowledge in HTN-planning, GOLOG-programming, and planning with temporal knowledge into ASP. To do so, we view domain-dependent control knowledge as sets of independent constraints. An advantage of this approach is that domain-dependent control knowledge can be modularly formalized and added to the planning problem as desired. We define a set of constructs for constraint representation and provide a set of domain-independent logic programming rules for checking constraint satisfaction.

This paper integrates research in robot programming and reasoning about action with research in model-based reasoning about physical systems to provide a capability for modeling and programming devices and web agents, which we term model-based programming. Model-based programs are reusable highlevel programs that capture the procedural knowledge of howto accomplish a task, without specifying all the device- and web-service-specific details. Model-based programs must be instantiated in the context of a model of a specific device/web service and state of the world. The instantiated programs are simply sequences of actions, which can be executed by an appropriate agent to control the behavior of the system. The separation of control and model enables reuse of model-based programs across classes of related devices and services whose configuration changes as the result of replacement, redesign, reconfiguration or component failure. Additionally, the logical formalism underlying model-based programming enables verification of properties such as safety, program existence, and goal achievement. Our model-based programs are realized by exploiting research on the logic programming language Golog, together with research on representing actions and state constraints in the situation calculus, and modeling physical systems using state constraints.

Suppose we are given a theory of system behavior and a set of candidate hypotheses. Our concern is with generating tests which will discriminate these hypotheses in some fashion. We logically characterize test generation as abductive reasoning. Aside from defining the theoretical principles underlying test generation, we are able to bring to bear the abundant research on abduction to show how test generation can be embodied in working systems. Furthermore, we address the issue of computational complexity. It has long been known that test generation is NP-complete. This is consistent with complexity results on the generation of abductive explanations. By syntactically restricting the description of our theory of system behavior or by limiting the completeness of our abductive reasoning, we are able to gain insight into tractable test generation problems.

This paper describes a qualitative technique for interpreting graphical data. Given a set of numerical observations regarding the behaviour of a system, its attributes can be determined by plotting the data and qualitatively comparing the shape of the resulting graph with graphs of system behaviour models. Qualitative data modeling incorporates techniques from pattern recognition and qualitative reasoning to characterize observed data, generate hypothetical interpretations, and select models that best fit the shape of the data. Domain-specific knowledge may be used to substantiate or refute the likelihood of hypothesized interpretations. The basic data modeling technique is domain independent and is applicable to a wide range of problems. It is illustrated here in the context of a knowledge-based system for well test interpretation. Cet article decrit une technique qualitative d'interpretation de donnees graphiques. A partir d'un ensemble de donnees numeriques portant sur le comportement d'un systeme, il est possible de determiner ses attributs en presentant les donnees sous forme graphique, puis en effectuant une comparaison qualitative de la courbe resultante avec des graphiques provenant de modeles comportementaux de systemes. La modelisation qualitative fait appel a des techniques de reconnaissance des formes et de raisonnement qualitatif pour caracteriser les donnees observees, generer des interpretations hypothetiques et choisir des modeles parmi les plus representatifs du format des donnees. Il est possible de recourir a des connaissances specifiques du domaine pour justifier ou refuter les probabilites d'interpretations hypothetiques. La technique fondamentale de modelisation est independante du domaine et s'applique a un vaste eventail de problemes. Elle est illustree ici dans le contexte d'un systeme de connaissances pour l'interpretation de tests de puits.