We introduce an efficient method for the verification of ReLU-based feed-forward neural networks. We derive an automated procedure that exploits dependency relations between the ReLU nodes, thereby pruning the search tree that needs to be considered by MILP-based formulations of the verification problem. We augment the resulting algorithm with methods for input domain splitting and symbolic interval propagation. We present Venus, the resulting verification toolkit, and evaluate it on the ACAS collision avoidance networks and models trained on the MNIST and CIFAR-10 datasets. The experimental results obtained indicate considerable gains over the present state-of-the-art tools.

We consider conjunctive query inseparability of description logic knowledge bases with respect to a given signaturea fundamental problem in knowledge base versioning, module extraction, forgetting and knowledge exchange. We give a uniform game-theoretic characterisation of knowledge base conjunctive query inseparability and develop worstcase optimal decision algorithms for fragments of Horn-ALCHI, including the description logics underpinning OWL 2 QL and OWL 2 EL. We also determine the data and combined complexity of deciding query inseparability. While query inseparability for all of these logics is P-complete for data complexity, the combined complexity ranges from P-to ExpTime-to 2ExpTime-completeness. We use these results to resolve two major open problems for OWL 2 QL by showing that TBox query inseparability and the membership problem for universal conjunctive query solutions in knowledge exchange are both ExpTime-complete for combined complexity. Finally, we introduce a more flexible notion of inseparability which compares answers to conjunctive queries in a given signature over a given set of individuals. In this case, checking query inseparability becomes NP-complete for data complexity, but the ExpTimeand 2ExpTime-completeness combined complexity results are preserved.

Ontop is a popular open-source virtual knowledge graph system that can expose heterogeneous data sources as a unified knowledge graph. Ontop has been widely used in a variety of research and industrial projects. In this paper, we describe the challenges, design choices, new features of the latest release of Ontop v4, summarizing the development efforts of the last 4 years.

No abstract

We investigate the problem whether two ALC ontologies are indistinguishable (or inseparable) by means of queries in a given signature, which is fundamental for ontology engineering tasks such as ontology versioning, modularisation, update, and forgetting. We consider both knowledge base (KB) and TBox inseparability. For KBs, we give modeltheoretic criteria in terms of (finite partial) homomorphisms and products and prove that this problem is undecidable for conjunctive queries (CQs), but 2ExpTime-complete for unions of CQs (UCQs). The same results hold if (U)CQs are replaced by rooted (U)CQs, where every variable is connected to an answer variable. We also show that inseparability by CQs is still undecidable if one KB is given in the lightweight DL EL and if no restrictions are imposed on the signature of the CQs. We also consider the problem whether two ALC TBoxes give the same answers to any query over any ABox in a given signature and show that, for CQs, this problem is undecidable, too. We then develop model-theoretic criteria for HornALC TBoxes and show using tree automata that, in contrast, inseparability becomes decidable and 2ExpTime-complete, even ExpTime-complete when restricted to (unions of) rooted CQs. module is extracted from an ontology. In ontology update or revision, the difference between the answers to queries over the updated or revised ontology and the original one should be minimised when constructing update or revision operators. Similarly, in forgetting, it is the answers to queries which should be preserved under appropriate forgetting operators. Thus, in the context of query answering, the fundamental relationship between ontologies is not whether they are logically equivalent (have the same models), but whether they give the same answers to any relevant query. To illustrate, consider the following simple TBox T = {Book ∃author.¬Book} saying that every book has an author who is not a book. Clearly, T is not logically equivalent to the TBoxwhich only states that every book has an author. However, if one takes as the query language the popular classes of conjunctive queries (CQs) or unions of CQs (UCQs), then no matter what the data is, every query will have the same answers independently of whether one uses T or T . Intuitively, the reason is that the 'positive' information given by T coincides with the 'positive' information given by T . If the main purpose of the ontology is answering UCQs, it is thus more important to know that T can be safely replaced by T without affecting the answers to UCQs than to establish that T and T are not logically equivalent.In most ontology engineering applications for ontology-based data access, the relevant class Q of queries can be further restricted to those given in a finite signature of relevant concept and role names. For example, to establish that a subset M of an ontology O is a module of O, one should not require that M and O give the same answers to all queries in Q, but only to those that are in the signature of M. Similarly, in the versioning ...