Automatically Created Concept Graphs Using Descriptive Keywords in the Medical Domain

Objectives: Besides keyword search, navigational search is an important means to find relevant information in digital object collections. Such navigation is often supported by categorization systems or thesauri, which provide a hierarchical view on a particular domain and allow for browsing digital collections. Existing categorization systems, however, require large and expensive efforts for the manual creation and maintenance. Our Semantic GrowBag algorithm fully automatically creates concept graphs, i.e. directed graphs similar to categorization systems but without strong subsumption semantics. This article sketches our algorithm and evaluates it for the medical domain.

Methods: Our Semantic GrowBag algorithm uses descriptive keywords and exploits higher-order cooccurrences between them to create concept graphs (so-called GrowBag graphs) from annotated object collections. In this study, we have automatically created more than 2000 GrowBag graphs based on the Medline data set to show the applicability of our algorithm in the medical domain. For the evaluation, we first compared our algorithm to a baseline algorithm that does not take higher-order co-occurrences into account, and then compared the resulting GrowBag graphs systematically against the manually crafted MeSH thesaurus.

Results: Our experiments revealed that the Semantic GrowBag approach essentially increases the number of relevant relationships in comparison to a baseline approach by about 50%. Furthermore, the identified relations usually correspond to and hardly ever contradict to relationships as stated by MeSH.

Conclusions: The Semantic GrowBag algorithm allows creating concept graphs fully automatically. While it does not systematically exploit specifics of a domain (such as the fundamental separation between ‘drugs’ and ‘therapy’ in MeSH), the resulting GrowBag graphs are nevertheless well-suited to support navigation in digital object collections. Moreover, they can also be used to help maintaining existing categorization systems based on the actual usage of categories.

Plasma amino and keto acids in chronic renal failure

During both early and late stages of chronic renal insufficiency the response of BCKA to the disease state, as indicated by plasma levels, differs from that of BCAA. Val is the only BCAA whose concentration changes under the conditions of our study, and this only during the more advanced stages of disease. In contrast, all three BCKA declined, KIVA and KICA even in mild renal failure, showing that already during the early stages of the disease these BCKA levels are decreased. BCKA are more sensitive parameters than the corresponding amino acids with regard to the metabolic dysfunctions characteristic of this disease. Modern analytical methods allow more exact and reliable knowledge of these indicators and thus a better understanding of biochemical mechanisms, possibly resulting in better therapy.

The truth will come to light: directions and challenges in extracting the knowledge embedded within trained artificial neural networks

To date, the preponderance of techniques for eliciting the knowledge embedded in trained artificial neural networks (ANN's) has focused primarily on extracting rule-based explanations from feedforward ANN's. The ADT taxonomy for categorizing such techniques was proposed in 1995 to provide a basis for the systematic comparison of the different approaches. This paper shows that not only is this taxonomy applicable to a cross section of current techniques for extracting rules from trained feedforward ANN's but also how the taxonomy can be adapted and extended to embrace a broader range of ANN types (e.g., recurrent neural networks) and explanation structures. In addition the paper identifies some of the key research questions in extracting the knowledge embedded within ANN's including the need for the formulation of a consistent theoretical basis for what has been, until recently, a disparate collection of empirical results.

Recruitment learning of boolean functions in sparse random networks

This work presents a new class of neural network models constrained by biological levels of sparsity and weight-precision, and employing only local weight updates. Concept learning is accomplished through the rapid recruitment of existing network knowledge - complex knowledge being realised as a combination of existing basis concepts. Prior network knowledge is here obtained through the random generation of feedforward networks, with the resulting concept library tailored through distributional bias to suit a particular target class. Learning is exclusively local - through supervised Hebbian and Winnow updates - avoiding the necessity for backpropagation of error and allowing remarkably rapid learning. The approach is demonstrated upon concepts of varying difficulty, culminating in the well-known Monks and LED benchmark problems.