Machine intelligence capable of learning complex procedural behavior, inducing (latent) programs, and reasoning with these programs is a key to solving artificial intelligence. The problems of learning procedural behavior and program induction have been studied from different perspectives in many computer science fields such as program synthesis [1], probabilistic programming [2], inductive logic programming [3], reinforcement learning [4], and recently in deep learning. However, despite the common goal, there seems to be little communication and collaboration between the different fields focused on this problem. Recently, there have been many success stories in the deep learning community related to learning neural networks capable of using trainable memory abstractions. This has led to the development of neural networks with differentiable data structures such as Neural Turing Machines [5], Memory Networks [6], Neural Stacks [7, 8], and Hierarchical Attentive Memory [11], among others. Simultaneously, neural program induction models like Neural Program-Interpreters [9] and the Neural Programmer [10] have created much excitement in the field, promising induction of algorithmic behavior, and enabling inclusion of programming languages in the processes of execution and induction, while remaining trainable end-to-end. Trainable program induction models have the potential to make a substantial impact on many problems involving long-term memory, reasoning, and procedural execution, such as question answering, dialog, and robotics. The aim of the NAMPI workshop is to bring together researchers and practitioners from both academia and industry, in the areas of deep learning, program synthesis, probabilistic programming, inductive programming and reinforcement learning, to exchange ideas on the future of program induction with a special focus on neural network models and abstract machines. Through this workshop we look to identify common challenges, exchange ideas and lessons learned from the different fields, as well as establish a (set of) standard evaluation benchmark(s) for approaches that learn with abstraction and/or reason with induced programs.

The Event Calculus is a temporal logic that has been used as a basis in event recognition applications, providing among others, direct connections to machine learning, via Inductive Logic Programming (ILP). We present an ILP system for online learning of Event Calculus theories. To allow for a single-pass learning strategy, we use the Hoeffding bound for evaluating clauses on a subset of the input stream. We employ a decoupling scheme of the Event Calculus axioms during the learning process, that allows to learn each clause in isolation. Moreover, we use abductive-inductive logic programming techniques to handle unobserved target predicates. We evaluate our approach on an activity recognition application and compare it to a number of batch learning techniques. We obtain results of comparable predicative accuracy with significant speed-ups in training time. We also outperform hand-crafted rules and match the performance of a sound incremental learner that can only operate on noise-free datasets. This paper is under consideration for acceptance in TPLP.

Artificial intelligence (AI) is intelligence exhibited by machines. In computer science, an ideal "intelligent" machine is a flexible rational agent that perceives its environment and takes actions that maximize its chance of success at some goal.[1] Colloquially, the term "artificial intelligence" is applied when a machine mimics "cognitive" functions that humans associate with other human minds, such as "learning" and "problem solving".[2] As machines become increasingly capable, facilities once thought to require intelligence are removed from the definition. For example, optical character recognition is no longer perceived as an exemplar of "artificial intelligence" having become a routine technology.[3] Capabilities still classified as AI include advanced Chess and Go systems and self-driving cars. AI research is divided into subfields[4] that focus on specific problems or on specific approaches or on the use of a particular tool or towards satisfying particular applications. The central problems (or goals) of AI research include reasoning, knowledge, planning, learning, natural language processing (communication), perception and the ability to move and manipulate objects.[5] General intelligence is among the field's long-term goals.[6] Approaches include statistical methods, computational intelligence, soft computing (e.g. machine learning), and traditional symbolic AI. Many tools are used in AI, including versions of search and mathematical optimization, logic, methods based on probability and economics. The AI field draws upon computer science, mathematics, psychology, linguistics, philosophy, neuroscience and artificial psychology. The field was founded on the claim that human intelligence "can be so precisely described that a machine can be made to simulate it."[7] This raises philosophical arguments about the nature of the mind and the ethics of creating artificial beings endowed with human-like intelligence, issues which have been explored by myth, fiction and philosophy since antiquity.[8] Attempts to create artificial intelligence has experienced many setbacks, including the ALPAC report of 1966, the abandonment of perceptrons in 1970, the Lighthill Report of 1973 and the collapse of the Lisp machine market in 1987. In the twenty-first century AI techniques became an essential part of the technology industry, helping to solve many challenging problems in computer science.[9]

One purpose of CASC is to provide a public evaluation of the relative capabilities of ATP systems. The TPTP version used for CASC is released beyond the ATP community. Fulfillment of these after the competition, so that new problems have not objectives provides insight and stimulus for the been seen by the entrants. In some divisions the systems development of more powerful ATP systems, leading are ranked according to the number of problems to increased and more effective use. The most recent CASC, accompanied by a proof or model (thus giving only held at CADE-25 in Berlin, Germany, in 2015, was an assurance of the existence of a proof/model).

A matched formula is a CNF formula whose incidence graph admits a matching which matches a distinct variable to every clause. Such a formula is always satisfiable. Matched formulas are used, for example, in the area of parametrized complexity. We prove that the problem of counting the number of the models (satisfying assignments) of a matched formula is #P-complete. On the other hand, we define a class of formulas generalizing the matched formulas and prove that for a formula in this class one can choose in polynomial time a variable suitable for splitting the tree for the search of the models of the formula. As a consequence, the models of a formula from this class, in particular of any matched formula, can be generated sequentially with a delay polynomial in the size of the input. On the other hand, we prove that this task cannot be performed efficiently for linearly satisfiable formulas, which is a generalization of matched formulas containing the class considered above.

In the future, a new generation of autonomous robots is set to complete tasks autonomously, even if something unforeseeable happens. With the support of the Austrian Science Fund FWF, information technology experts in Graz are working to advance the development of artificial intelligence and equip robots with common sense. Something that children learn through play and that adults are able to do on the basis of past experience, such as responding to unexpected situations, remains one of the great challenges in robotics. Autonomous systems are expected to complete tasks given to them without external input. The deployment of such intelligent robots would be particularly important in critical situations – such as environmental disasters or industrial accidents.

In the future, a new generation of autonomous robots is set to complete tasks autonomously, even if something unforeseeable happens. With the support of the Austrian Science Fund FWF, information technology experts in Graz are working to advance the development of artificial intelligence and equip robots with common sense. Something that children learn through play and that adults are able to do on the basis of past experience, such as responding to unexpected situations, remains one of the great challenges in robotics. Autonomous systems are expected to complete tasks given to them without external input. The deployment of such intelligent robots would be particularly important in critical situations – such as environmental disasters or industrial accidents.

Many representation schemes combining first-order logic and probability have been proposed in recent years. Progress in unifying logical and probabilistic inference has been slower. Existing methods are mainly variants of lifted variable elimination and belief propagation, neither of which take logical structure into account. We propose the first method that has the full power of both graphical model inference and first-order theorem proving (in finite domains with Herbrand interpretations). We first define probabilistic theorem proving (PTP), their generalization, as the problem of computing the probability of a logical formula given the probabilities or weights of a set of formulas.

A goal of research in artificial intelligence and machine learning since the early days of expert systems has been to develop automated reasoning methods that combine logic and probability. Why is there a need to combine logic and probability? Probability theory allows one to quantify uncertainty over a set of propositions--ground facts about the world--and a probabilistic reasoning system allows one to infer the probability of unknown (hidden) propositions conditioned on the knowledge of other propositions. However, probability theory alone has nothing to say about how propositions are constructed from relationships over entities or tuples of entities, and how general knowledge at the level of relationships is to be represented and applied.

This paper studies two issues concerning relevance in structured argumentation in the context of the ASPIC+ framework, arising from the combined use of strict and defeasible inference rules. One issue arises if the strict inference rules correspond to classical logic. A longstanding problem is how the trivialising effect of the classical Ex Falso principle can be avoided while satisfying consistency and closure postulates. In this paper, this problem is solved by disallowing chaining of strict rules, resulting in a variant of the ASPIC+ framework called ASPIC*, and then disallowing the application of strict rules to inconsistent sets of formulas. Thus in effect Rescher & Manor's paraconsistent notion of weak consequence is embedded in ASPIC*. Another issue is minimality of arguments. If arguments can apply defeasible inference rules, then they cannot be required to have subset-minimal premises, since defeasible rules based on more information may well make an argument stronger. In this paper instead minimality is required of applications of strict rules throughout an argument. It is shown that under some plausible assumptions this does not affect the set of conclusions. In addition, circular arguments are in the new ASPIC* framework excluded in a way that satisfies closure and consistency postulates and that generates finitary argumentation frameworks if the knowledge base and set of defeasible rules are finite. For the latter result the exclusion of chaining of strict rules is essential. Finally, the combined results of this paper are shown to be a proper extension of classical-logic argumentation with preferences and defeasible rules.