With the recent emergence of the AI black box problem, the fairness, accountability, and transparency (FAT) of AI have been vigorously debated. Companies and academic institutions have developed various types of explainable AI. Fujitsu Laboratories has developed a new technology, Wide Learning, which is not only capable of dealing with the black box problem but also includes explainability that achieves knowledge discovery. Wide Learning uses "enumeration," which is a major technology of discovery science, to exhaustively enumerate pieces of knowledge called knowledge chunks described in a human-understandable format. By using this knowledge for judgment, the overlooking of potential candidates is significantly reduced and high-accuracy prediction and classification are achieved. If knowledge without omission can be discovered continuously, the trust of service systems can be strengthened and knowledge diversion in turn should provide trust between service systems. This paper first outlines the history of how Wide Learning came into being and introduces its technical characteristics. It then reports on a demonstration of knowledge discovery using Wide Learning in Fujitsu and presents future prospects.
Systems of engagement (SoE) are ICT systems for creating and maintaining connections between companies and customers and building trust. SoE must be able to offer flexibility and quickness in responding to changes in business, in addition to availability comparable to that of systems of record (SoR), which are ICT systems that support the mission-critical operations in an organization. For services provided by SoE to be used by more users, a system must essentially satisfy the security and integrity requirements that allow for a greater sense of security in users. However, grasping the configuration and requirements of an entire system is generally difficult for service developers, and it was difficult to redesign a system considering the security and performance based on the service every time improvements were made to the service. Therefore, Fujitsu Laboratories has developed a system of building SoE and supporting stable operations so that such service developers can quickly and flexibly provide a secure service in a multi-cloud environment. This paper presents network as code (NaC) and network verification technologies that assist even those service developers who lack an understanding of the entire system with the construction of a secure system. It also describes traffic prediction and anomaly detection technologies, which promptly recognize any failures and performance degradation in an SoE system to allow for the provision of integrity.
Blockchain is a technology devised as the fundamental technology of Bitcoin. It allows reliability to be ensured in a decentralized manner, and expectations are high for its application in various areas beyond virtual currency, such as real estate and healthcare. However, any problem in the boundaries between the elements that constitute an entire system or in the smart contracts that are often introduced as subsystems may lead directly to significant losses in business, such as the theft of the virtual currency managed by a blockchain. Accordingly, improving the reliability of the system as a whole, including applications, is needed. Fujitsu Laboratories has developed a threat analysis method that checks a blockchain system for any problems at the time of its construction and operation, and a smart contract verification technology that exhaustively detects threats in smart contracts by using static analysis technology. These will enable blockchain developers to quickly develop systems that use blockchains with greater security. This paper describes the threat analysis method and smart contract verification technology that we have developed.
In order to realize and promote a cashless society, payment using only the biometric information of an individual without the need for identification by ID card or smartphone is attracting attention. In cases where authentication on a scale of one million users is assumed, such as payments at brick-and-mortar stores and admission to event venues, it is necessary to search for and authenticate a single individual from a huge volume of personal data. However, high-speed searches of large volumes of data were difficult with one biometric modality alone. For that reason, inputting additional information such as IDs and personal identification numbers (PINS) was necessary, which compromised user convenience. Accordingly, Fujitsu Laboratories has developed an integrated biometric authentication technology that combines palm vein and face recognition. This technology uses facial images effortlessly captured during payment terminal use to narrow down the possible matches, thereby allowing the number of system users to be increased without the need for the input of information other than biometric data. This paper describes the present state and issues of payments using biometric authentication, and the integrated biometric authentication technology developed.
In recent years, business and lifestyles have been undergoing significant changes with the progress of networks and ICT. Fundraising and information provision, conventionally difficult for individuals, have become easier with the use of affiliate marketing, crowdfunding, websites, social media, and so on. In addition, the emergence of the sharing economy has made various services available at low cost. As this trend becomes more widespread, however, problems between the people involved in transactions and communication on the Internet are emerging. In response to this issue, Fujitsu Laboratories has developed an !CT risk assessment technology for quantifying users' ICT risks based on the psychological and behavioral characteristics observed in people who have experienced harm from cyber attacks, as well as a reliability scoring technology for expanding the scope of application of this assessment technology to various other risks besides ICT risks. These technologies allow for the scoring of a variety of risks, including delays and inadequate service provision even for people with whom transactions are conducted for the first time. The scores obtained in this way can be utilized to develop and implement measures aimed at preventing problems. This paper describes these ICT risk assessment and reliability scoring technologies as well as examples of their application.
