There are two main approaches to the development of AI systems : data-driven approaches primarily using machine learning and knowledge-based approaches using knowledge processing technologies. In knowledge-based AI, various knowledge graphs are developed and used as core technologies. A knowledge graph represents connections between various pieces of knowledge in a network-like graph structure, and technologies such as Linked Data and ontologies are widely used for its publication and utilization. In this paper, the foundational technologies involved in the development of AI systems using knowledge graphs are explained.

It is well known that “nonlinear partial differential equations,” which are typified by mathematical models for transforming objects such as strings, beams, and films, Burgers equations in fluid mechanics, and Schrödinger equations in quantum mechanics, can represent a lot of phenomena in the world of nature. However, it is quite difficult to control such systems because of nonlinearity and infinite dimensionality; hence, they are actively studied by various approaches. There are some approaches, called “discrete mechanics” and “ultradiscretization,” based on discrete structures of nonlinear partial differential equations, and it corresponds to the extraction of the “discrete bony framework” from the systems. Moreover, we can expect control synthesis methods that are highly compatible with computers treating only digital data. This paper shall provide some fundamental and applied research studies on this topic with our group's results.

Deep neural networks have achieved remarkable success in image classification tasks, but their performance is known to be vulnerable to incorrect labels, also known as noisy labels. In this paper, we provide a comprehensive overview of robust learning methods designed to mitigate the impact of noisy labels. We first define the concept of noisy labels and discuss the different patterns of label noise. We then introduce real-world datasets that contain noisy labels and outline a promising approach to semisupervised learning for improving robustness against label noise. Finally, we explore the details of methods that enhance robustness to noisy labels, including regularization, label correction, sample selection, and the combination of sample selection with semisupervised learning.

Recently, an optical information processing infrastructure has been developed to achieve low-latency, high-capacity communication and high-speed, high-performance information processing with low power consumption. Security functions are required to implement using an optical technology for the infrastructure. Therefore, a study has begun to be proceeded. In this article, we introduce photonic cryptographic circuits using photonic logic gates to construct a nonlinear function of block ciphers. We present silicon photonic chips of the cryptographic function and their experimental evaluations. We also present an overview of the realization of cryptographic circuits using the optical technology.