Modular structurality and emergent functionality within knowledge representation systems

Semina Nr 15 Scientiarum 2016 s. 77–87 DOI: http://dx.doi.org/10.15633/ss.1769 Adam Fedyniuk Modular structurality and emergent functionality within knowledge representation systems In contemporary science, at the crossroads between cognitive sci‑ ence and knowledge engineering one can find a research niche that can benefit substantially from interdisciplinary work. The discipline that concerns this particular area of scientific inquiry, and is most‑ ly relevant to the subject of this paper is known as ontology meta‑ modelling. Ontology metamodelling is a discipline that uses vari‑ ous tools to propose and test on different levels the viability of new and innovative approaches and architectures that may be applied to knowledge engineering. As is commonly stated, in ontology design there is no one right and true way of choosing the adequate tools and methodology. Some ideas are worthy of being tested in practical terms instantly when they are devised (due to level their level of de‑ tail and specificity), but some need a thorough analysis and strong conceptual framework as it’s foundations to even be considered vi‑ able. That is one of the main goals of metamodelling in general: to check if those bold ideas, and inspirations can have a meaningful application as well as coherent theoretical structure. The necessity for it stems from the fact, that oftentimes some explanations and ap‑ proaches in knowledge engineering may be too resource‑dependent and in the end, time‑consuming. In some ways this discipline may be considered akin to the practical considerations within the dis‑ course of philosophy of information, and in many ways there can 78 Adam Fedyniuk be seen many similarities in the areas of expertise. Yet, as it may seem, the philosophical approach emphasizes interdisciplinarity, and at times very strict scrutiny that may be applied to the theo‑ retical framework of analysed solutions. 1. Connection between interdisciplinary domains One of the major areas where one can find an inspiration for new ideas concerning the structurality found in artificial knowledge rep‑ resentation systems (such as domain ontologies in conjunction with their editors equipped with various reasoning engines) can be seen in biology.1 To further emphasize this connection, it is worthwhile to look at similarities between connectomics and proteomics. Both of these disciplines delve into the structurality of information pro‑ cessing, and both employ the same mathematical theories for the analysis of it’s structures. Especially one seems to be viable in wide variety of subjects in the interdisciplinary work – centrality. A math‑ ematical theory that is widely used in social network analysis, was proven useful both in the fields of proteomics and connectomics.2 Additionally, the structure of the content within various ontologies can be represented in network or graph format and that makes it also a subject to network analysis and it’s structure prone to cen‑ trality measurement.3 With network structure in mind, it is far easier to consider cre‑ ating a modular architecture for artificial knowledge representa‑ tion systems. In many cases, such ideas are explored, but one must consider another side that contributes to innovative ontology de‑ F. Azam, Biologically inspired modular neural networks, Blacksburg, VA 2000. F. Cheng and others, Quantitative network mapping of the human kinome interactome reveals new clues for rational kinase inhibitor discovery and individualized cancer therapy, “Oncotarget” 15 (2014), p. 3697–3710, doi: 10.18632/oncotarget. 1984; Zuo X. N., R and others, Network centrality in the human functional connectome, “Cerebral Cortex” 22 (2012) pp. 1862–1875, doi: 10.1093/cercor/bhr269. 3 C. Roche, Network analysis of Semantic Web Ontologies, Stanford CS224W Social and Information Network Analysis 2011. 1 2 Modular structurality and emergent functionality… 79 sign – the biological structures that could be the meaningful inspi‑ ration for new approaches and solutions in knowledge engineering. In some way one can protest with saying that it contributes to reduc‑ tionist claims and overall can be limiting for ontology designer, but it is possible that within biological systems we can find bizzare and hard to explain phenomena that may in future become paramount to meaningful application of ontologies. One of such phenomena, that is highly elusive and provoked a lot of discord between philos‑ ophers is the mechanism of emergence. In systems biology emer‑ gent functions can be found within various metabolic tracts. The flow of information between different protein domains creates a ki‑ nase network that exhibit emergent properties and functions, and as such they do so by maintaining a modular structure, that chang‑ es it’s pathways in response to damage and maintains the stability of information flow, sometimes by being supported by parts of the network that normally takes part in enabling completely different function. The modular structure of those networks clearly is an ap‑ propriate foundation for those emergent properties. One can infer that if we are able to duplicate this effect when considering the fea‑ tures of modular ontologies, it would be a great step forward in op‑ timisation of ontology mapping and creating a new level of expres‑ sion those systems would work on. 2. The extension of modularity application in novel approaches There are proposals that employ modularity and knowledge rep‑ resentation concepts into artificial neural network.4 Moreover, the notion of interweaving neural networks and knowledge representa‑ tion systems was proposed in regard to ontology mapping.5 We have 4 I. Kollia and others, Interweaving knowledge representation and adaptive neural networks, “Workshop on Inductive Reasoning and Machine Learning on the Se‑ mantic Web” 2009. 5 Y. Peng, Ontology mapping neural network: an approach to learning and inferring correspondences among ontologies, Pittsburgh 2010. 80 Adam Fedyniuk yet to see a proposal that would employ any form of modularity in a meaningful way into those two venues of scientific discourse. Both of mentioned claims are concerned with hybrid structure, where the architecture of the system is very complex. A question addressing this complexity can arise: would such systems benefit from modu‑ lar structure? In the response to interweaving knowledge represen‑ tation system and ANN the answer is yes, but in regard to the sec‑ ond claim, the question remains open. Although when we consider the possibility of utilizing emergent functions, then both of those claims would certainly benefit from it, because of the economy of function implementation that would be inherent and solely caused by structurality of the system itself. There is another issue at stake here: to what extent we can depend on the stability of emergent func‑ tions in artificial systems? The all‑encompassing ability to autoreg‑ ulate features of biological (...truncated)