Generative large language models (LLMs) have transformed AI by enabling rapid, human-like text generation, but they face challenges, including managing inaccurate information generation. Strategies such as prompt engineering, Retrieval-Augmented Generation (RAG), and incorporating domain-specific Knowledge Graphs (KGs) aim to address their issues. However, challenges remain in achieving the desired levels of management, repeatability, and verification of experiments, especially for developers using closed-access LLMs via web APIs, complicating integration with external tools. To tackle this, we are exploring a software architecture to enhance LLM workflows by prioritizing flexibility and traceability while promoting more accurate and explainable outputs. We describe our approach and provide a nutrition case study demonstrating its ability to integrate LLMs with RAG and KGs for more robust AI solutions.

Analysts who are interested in quickly identifying new and emerging scientific advancements have numerous challenges as the breadth, depth, and volume of scientific literature increases. Network analysis and mining is key to the success in this task. The ARBITER system seeks to identify indicators of emergence and provide a system that is capable of analyzing corpora of full text and metadata to identify emerging science topics and explain its reasoning and conclusions. In this paper, we describe a network-modeling framework that is used in the ARBITER system, and describe our novel hybrid approach using probabilistic foundations in combination with semantic technology and introduce our explanation infrastructure. We include a discussion of some challenges and opportunities related to explaining hybrid approaches to indicator-based analysis and emergence detection.

We present a novel ensemble architecture for learning problem-solving techniques from a very small number of expert solutions and demonstrate its effectiveness in a complex real-world domain. The key feature of our “Generalized Integrated Learning Architecture” (GILA) is a set of heterogeneous independent learning and reasoning (ILR) components, coordinated by a central meta-reasoning executive (MRE). The ILRs are weakly coupled in the sense that all coordination during learning and performance happens through the MRE. Each ILR learns independently from a small number of expert demonstrations of a complex task. During performance, each ILR proposes partial solutions to subproblems posed by the MRE, which are then selected from and pieced together by the MRE to produce a complete solution. The heterogeneity of the learner-reasoners allows both learning and problem solving to be more effective because their abilities and biases are complementary and synergistic. We describe the application of this novel learning and problem solving architecture to the domain of airspace management, where multiple requests for the use of airspaces need to be deconflicted, reconciled, and managed automatically. Formal evaluations show that our system performs as well as or better than humans after learning from the same training data. Furthermore, GILA outperforms any individual ILR run in isolation, thus demonstrating the power of the ensemble architecture for learning and problem solving.

The Open Government Directive is making US government data available via websites such as Data.gov for public access. In this paper, we present a Semantic Web based approach that incrementally generates Linked Government Data (LGD) for the US government. In focusing on the trade-off between high quality LGD generation (requiring non-trivial human expert input) and massive LGD generation (requiring low human processing cost), our work is highlighted by the following features: (i) supporting low-cost and extensible LGD publishing for massive government data; (ii) using Social Semantic Web (Web3.0) technologies to incrementally enhance published LGD via crowdsourcing, and (iii) facilitating mash-ups by declaratively reusing cross-dataset mappings which usually are hard-coded in applications.

Many emerging semantic web applications include ontologies from one set of authors and instance data from another (often much larger) set of authors. Often ontologies are reused and instance data is integrated in manners unanticipated by their authors. Not surprisingly, many instance data rich applications encounter instance data that is not compatible with the expectations of the original ontology author(s). This line of work focuses on issues related to semantic expectation mismatches in instance data. Our initial results include a customizable and extensible service-oriented evaluation architecture, and a domain implementation called PmlValidator, which checks instance data using the corresponding ontologies and additional style requirements.

Corporate databases increasingly are being viewed as potentially rich sources of new and valuable knowledge. Various approaches to “discovering” or “mining” such knowledge have been proposed. Here we identify an important and previously ignored discovery task, which we call data archaeology. Data archaeology is a skilled human task, in which the knowledge sought depends on the goals of the analyst, cannot be specified in advance, and emerges only through an iterative process of data segmentation and analysis. We describe a system that supports the data archaeologist with a natural, object-oriented representation of an application domain, a powerful query language and database translation routines, and an easy-to-use and flexible user interface that supports interactive exploration. A formal knowledge representation system provides the core technology that facilitates database integration, querying, and the reuse of queries and query results.

CLASSIC is a data model that encourages the description of objects not only in terms of their relations to other known objects, but in terms of a level of intensional structure as well. The CLASSIC language of structured descriptions permits i) partial descriptions of individuals, under an 'open world' assumption, ii) answers to queries either as extensional lists of values or as descriptions that necessarily hold of all possible answers, and iii) an easily extensible schema, which can be accessed uniformly with the data. One of the strengths of the approach is that the same language plays multiple roles in the processes of defining and populating the DB, as well as querying and answering.

Knowledge Representation (KR) systems provide support for Artificial Intelligence systems that reason about relationships between objects in their domains of expertise. Because of their support for inference, KR systems appear to have potential to enrich the kind of retrievals that IR systems might make. Ironically, however, the most useful KR systems are limited to reasoning based on a rigid notion of validity, and thus are awkward to use when relevant but inexact retrievals are desired. We have been exploring the potential of a “connectionist” model—the Boltzmann Machine—to overcome this limitation. We report on a number of experiments in which we use a connectionist simulator to support similarity-based reasoning in a frame representation. We draw some tentative, mixed conclusions on the potential for a union of KR, IR, and connectionism.