Fujitsu's Latest Development in Machine Learning Technology Enhances Accuracy in Graph Structured Data
KAWASAKI, JAPAN: Fujitsu’s latest development in machine learning technology transforms various unstructured data into a uniform expression called “tensor” to achieve high-accuracy in graph-structured data. The new development with a graph structure expresses the relationships between people and things.
“Tensor” is a generalized concept of vectors and matrices used to represent multidimensional arrays in data.
The Graph-structured data comes with a finite set of nodes along with a set of unordered pairs for an undirected graph. It has a complicated structure with variability in size and methods of expression.
Graph-structured Data Architecture
The data are transformed into a uniform expression format using a mathematical operation called “tensor factorization”. The new technology from Fujitsu can perform tensor factorization to maximize the degree of similarity to an arbitrary pattern. It further reduces the uniform expressions to maximize the accuracy of categorization. “Tensor factorization” is a technology that factors multidimensional arrays based on the sum of correlations between multiple elements.
The technology finds application in learning the structure and activities of chemical compounds based on data from the PubChem BioAssay open database of chemical compounds. The new offering increases the accuracy predictability of compounds compared to the existing technology. The technology can also be employed for financial transactions and chemical compositions. It is further capable of detecting the malicious threats deploying support vector machines to increase the efficiency of network monitoring tasks.