The convolutional neural network using a large number of images with ground truth had been the defacto standard in image recognition since its overwhelming victory at ILSVRC2012. However, in recent years, unsupervised representation learning and Transformer, which is a different approach from CNN, have been proposed, and image recognition based on deep learning is progressing further. In this paper, I introduce recent research trends in image recognition with a focus on unsupervised representation learning and Transformer.

Environmental sound analysis and anomalous sound detection have seen extensive development, and they are becoming an established field of acoustic and audio processing. In this review paper, we outline the research topics addressed in environmental sound analysis and anomalous sound detection, and we introduce the challenges and recent research trends, as well as our recent works.

Only humans are able to understand the meaning of images and complement them. Language seems to be deeply involved in the development of this ability. After the invention of writing, the cave paintings of ancient humans depicted not only real creatures such as bison and deer, but also imaginary creatures such as gods and spirits. This can be called symbolic pictures. A picture is better at evoking emotions than text. Analyzing and generating symbolic pictures with artificial intelligence will lead to the understanding of human emotion recognition. This kind of research method is called analysis and synthesis. By this method, I have created an artificial intelligence that can draw imaginary creatures and manga. Perhaps it is time to move the target from meanings generated by pictures to human emotions evoked by those meanings. Analysis and synthesis confirmed that picture is a “language” for expressing emotions. If it is natural that language has grammar, then the desire with full use of that grammar is the root of everything from cave paintings to Manga, and is the impression that evoke human deep emotions.

It has been pointed out that the coloring problem in graph theory can be applied to channel assignment in wireless communication, and numerous studies have been carried out. Initially, the study was conducted in the framework of the cellular system, assuming a car phone. Since then, wireless communication has advanced and a wide range of applications is being considered. Graph coloring is expected to be applicable to wireless communication, but it is not sufficient at present. Therefore, in this paper, we will introduce the past of channel assignment and graph coloring, which is expected to support advanced wireless communication in the future. Finally, we illustrate graph examples that can be applied to wireless sensor networks.

FUJITSU and RIKEN Center for Computational Science (R-CCS) was involved in the FLAGSHIP2020 project to develop a Japanese flagship supercomputer and successfully developed the supercomputer Fugaku as a successor to the K computer. The goal of the project was to develop a supercomputer system on which various practical applications run with high performance and reasonable power consumption. To achieve this goal, a unique framework called co-design, in which the system development team and the application development team closely cooperated, was adopted. For example, several decisions on the architecture of the Fugaku system were made by analyzing the characteristics of real applications through the co-design process. While the co-design activity covered a wide range of Fugaku development works from hardware to system software and applications, this paper will focus mainly on the development of the processor (A64FX) of Fugaku.

Decision making is important in information and communications technology and artificial intelligence. Here, we present photonics approaches to realize decision making using classical and quantum light. We demonstrate a single-photon-based system to solve multi-armed bandit problems by utilizing wave-particle duality for exploration. The principle is applied to chaotic lasers where the chaotically oscillating time series provides fast and scalable decision-making abilities. Furthermore, we demonstrate the usefulness of quantum interference effects in social decision-making, whereby the expected value regards maximization of social benefits and ensuring fairness among individuals. The impact of near-field optics for computing is also discussed.

A research program for studying ethical, legal, and social implications of science and technology research from mainly humanities and social science perspectives is called “ELSI.” This paper provides a historical overview of the emergence of these ideas and practices in the United States at the end of the last century and their subsequent development. This study will provide valuable suggestions about ELSI issues in the field of information technology.