Soonhyun Kwon

Augmented ontology by handshaking with machine learning

Artificial intelligence products are already around us and will be emerging dramatically a lot in near future. Artificial intelligence is all about data analysis. When it comes to data analysis, there are two representative techniques: machine learning and semantic technology. They stand on the other side from where to begin analysis. Simply speaking, machine learning is based on the data while semantic technology relies on human domain knowledge (human learning). What if collected data are insufficient to reflect whole phenomenon? This is a limitation of machine learning. What if circumstance changes a lot as time goes by? Manual rule updating by experts is not a good solution in that circumstance. Based on these observations, we investigate two approaches and find a good solution which maximizes the advantages of both techniques and mitigates the limitations of them. This paper suggests a novel integration idea to compensate each technology with the other: that is semantic filtering. This paper includes a toy semantic modelling and a machine learning algorithm implementation to realize the proposed concept, semantic filtering.

PAGE: Answering Graph Pattern Queries via Knowledge Graph Embedding.

Answering graph pattern queries have been highly dependent on a technique—i.e., subgraph matching, however, this approach is ineffective when knowledge graphs include incorrect or incomplete information. In this paper, we present a method called \(\mathtt {PAGE}\) that answers graph pattern queries via knowledge graph embedding methods. \(\mathtt {PAGE}\) computes the energy (or uncertainty) of candidate answers with the learned embeddings and chooses the lower-energy candidates as answers. Our method has the two advantages: (1) \(\mathtt {PAGE}\) is able to find latent answers hard to be found via subgraph matching and (2) presents a robust metric that enables us to compute the plausibility of an answer. In evaluations with two popular knowledge graphs, Freebase and NELL, \(\mathtt {PAGE}\) demonstrated the performance increase by up to 28% compared to baseline KGE methods.

A design of IoT device similarity vector based workflow management system

Nowadays, versatile IoT Devices are connected through the Internet and provide numerous dynamic services. However, until now, these kinds of services are just following established service or application rules. For this reason, an application cannot deal with a condition when rule is not previously set. Additionally, rules should be renewed by reflecting each condition, and software should be re-developed which are costly and time-consuming. To address this issue, we suggest a machine learning and semantic information based IoT devices management system.