In this extended abstract we examine the principles that underlie the construction of what we call Ontologies in Variation — a human-comprehensible knowledge representation scheme for natural kinds, objects, and concepts that captures both prototypical (or canonical) properties of classes of objects as well as those properties that are in variation. A fundamental characteristic of our work is that the variability captured in our representation is derived from data and as such that the provenance of statistical knowledge — the dataset — is directly associated with the ontology. This reliance on empirical data directs us towards a frequentist view of variation as statistical assertions, in contrast to much of the current work that integrates logic and uncertainty. Our formalism's novelty lies in the strategic complementation of axiomatic knowledge by statistical knowledge, and by the desire to preserve human comprehension of the resulting representation. We illustrate this work in the context of an ongoing project to create a representation of human anatomy — a queryable digital anatomy book that fits all of us in some variation.

This paper describes an end-to-end learning framework that allows a novice to create a model from data easily by helping structure the model building process and capturing extended aspects of domain knowledge. By treating the whole modeling process interactively and exploiting high-level knowledge in the form of an ontology, the framework is able to aid the user in a number of ways, including in helping to avoid pitfalls such as data dredging. Prudence must be exercised to avoid these hazards: certain conclusions may be supported by extra knowledge if, for example, there are reasons to trust a particular narrower set of hypotheses. This paper adopts the solution of using higher-level knowledge in order to allow this sort of domain knowledge to be inferred automatically, thereby selecting only relevant input attributes and thence constraining the hypothesis space. We describe how the framework automatically exploits structured knowledge in an ontology to identify relevant concepts, and how a data extraction component can make use of online data sources to find measurements of those concepts so that their relevance can be evaluated. To validate our approach, models of four different problem domains were built using our implementation of the framework. Prediction error on unseen examples of these models show that our framework, making use of the ontology, helps to improve model generalization.

ABox abduction plays an important role in reasoning over description logic (DL) ontologies. However, it does not work with inconsistent DL ontologies. To tackle this problem while achieving tractability, we generalize ABox abduction from the classical semantics to an inconsistency-tolerant semantics, namely the Intersection ABox Repair (IAR) semantics, and propose the notion of IAR-explanations in inconsistent DL ontologies. We show that computing all minimal IAR-explanations is tractable in data complexity for first-order rewritable ontologies. However, the computational method may still not be practical due to a possibly large number of minimal IAR-explanations. Hence we propose to use preference information to reduce the number of explanations to be computed.

Datalog-based systems often materialise all consequences of a datalog program and the data, allowing users' queries to be evaluated directly in the materialisation. This process, however, can be computationally intensive, so most systems update the materialisation incrementally when input data changes. We argue that existing solutions, such as the well-known Delete/Rederive (DRed) algorithm, can be inefficient in cases when facts have many alternate derivations. As a possible remedy, we propose a novel Backward/Forward (B/F) algorithm that tries to reduce the amount of work by a combination of backward and forward chaining. In our evaluation, the B/F algorithm was several orders of magnitude more efficient than the DRed algorithm on some inputs, and it was never significantly less efficient.

We design a tractable Horn fragment of the Halpern-Shoham temporal logic and extend it to interval-based temporal description logics, instance checking in which is P-complete for both combined and data complexity.

Topic modeling is a powerful tool for uncovering latent structure in many domains, including medicine, finance, and vision. The goals for the model vary depending on the application: sometimes the discovered topics are used for prediction or another downstream task. In other cases, the content of the topic may be of intrinsic scientific interest. Unfortunately, even when one uses modern sparse techniques, discovered topics are often difficult to interpret due to the high dimensionality of the underlying space. To improve topic interpretability, we introduce Graph-Sparse LDA, a hierarchical topic model that uses knowledge of relationships between words (e.g., as encoded by an ontology). In our model, topics are summarized by a few latent concept-words from the underlying graph that explain the observed words. Graph-Sparse LDA recovers sparse, interpretable summaries on two real-world biomedical datasets while matching state-of-the-art prediction performance.

Applications of description logics (DLs) such as ontology-based data access (OBDA) require understanding of how to pose database queries over DL knowledge bases. While there have been many studies regarding traditional relational query formalisms such as conjunctive queries and their extensions, little attention has been paid to graph database queries, despite the fact that graph databases have essentially the same structure as knowledge bases. In particular, not much is known about the interplay between DLs and XPath. The latter is a powerful formalism for querying semistructured data: it is in the core of most practical query languages for XML trees, and it is also gaining popularity in theory and practice of graph databases. In this paper we make a step towards coupling knowledge bases and graph databases by studying how to answer powerful XPath-style queries over simple DLs like DL-Lite and EL. We start with adapting the definition of XPath to the DL context, and then proceed to study the complexity of evaluating XPath queries over knowledge bases. Results show that, while query answering is undecidable for the full XPath, by carefully tuning the shape of negation allowed in the queries we can arrive at XPath fragments that have a potential to be used in practice.

Module extraction — the task of computing a (preferably small) fragment M of an ontology T that preserves entailments over a signature S — has found many applications in recent years. Extracting modules of minimal size is, however, computationally hard, and often algorithmically infeasible. Thus, practical techniques are based on approximations, where M provably captures the relevant entailments, but is not guaranteed to be minimal. Existing approximations, however, ensure that M preserves all second-order entailments of T w.r.t. S, which is stronger than is required in many applications, and may lead to large modules in practice. In this paper we propose a novel approach in which module extraction is reduced to a reasoning problem in datalog. Our approach not only generalises existing approximations in an elegant way, but it can also be tailored to preserve only specific kinds of entailments, which allows us to extract significantly smaller modules. An evaluation on widely-used ontologies has shown very encouraging results.

Sorted Neighborhood is an established blocking method for relational databases. It has not been applied on graph-based data models such as the Resource Description Framework (RDF). This poster presents a modular workflow for applying Sorted Neighborhood to RDF. Real-world evaluations demonstrate the workflow's utility against a popular baseline. Entity Resolution (ER) is the abstract problem of identifying Figure 1: A simple instance of ER in an RDF graph pairs of entities across databases that are syntactically disparate but logically equivalent. The problem goes by multiple names in the AI community, examples being record Table 1: Tuples sorted using blocking key values (BKVs) linkage, instance matching, and coreference resolution (Elmagarmid, ID First Name Last Name Zip BKV Ipeirotis, and Verykios 2007).

With the current upward trend in semantically annotated data, ontology-based data access (OBDA) was formulated to tackle the problem of data integration and query answering, where an ontology is formalized as a description logic TBox. In order to meet usability requirements set by users, efforts have been made to equip OBDA system with explanation facilities. One important explanation tool for DL ontologies, referred to as query abduction, can be formalised as abductive reasoning. In particular, given an ontology and an observation (i.e., a query with an answer), an explanation to the observation is a set of facts that together with the ontology can entail the observation. In this paper, we develop a sound and complete algorithm of query abduction for general conjunctive queries in ELH ontologies. This is achieved through ontology approximation and query rewriting. We implemented a prototypical system using the highly optimized Prolog engine XSB. We evaluated our algorithm over university benchmark ontology and our experimental results show that the system is capable of handling query abduction problems for ontology that has approximately 10 millions ABox assertions.