Abstract Federated Learning (FL) enables multiple clients to collaboratively train a shared model without exposing local data, making it a fundamental paradigm for large-scale distributed intelligence. However, in practical edge-cloud deployments, the server must inspect a large volume of high-dimensional client updates within tight communication windows, which makes secure aggregation a problem closely tied to parallel processing, real-time response, and high-performance computing (HPC) resources. Among the major threats to FL, backdoor attacks are particularly insidious because they implant malicious behaviors into the global model while preserving benign-task performance. Although pre-aggregation defenses based on gradient analysis are promising, the current state-of-the-art methods such as Scope suffer from two key limitations: fixed parameters are ineffective against adaptive attackers, and single-point clustering is vulnerable to failure under heterogeneous (non-IID) data distributions. To address these limitations, we propose FAROS, a robust and HPC-friendly defense framework that generalizes the transform-and-cluster paradigm. FAROS incorporates two key components: Adaptive Differential Scaling (ADS), which dynamically adjusts defense sensitivity according to the dispersion of client gradients, and Robust Core-set Computing (RCC), which replaces single-point clustering with a consensus-based centroid derived from a stable core-set. This design improves robustness while preserving server-side efficiency through vectorizable similarity computation and parallelizable filtering. Extensive experiments on multiple datasets, models, and attack settings show that FAROS consistently outperforms existing defenses in both attack suppression and benign-task accuracy, while remaining compatible with scalable distributed FL infrastructures.

In recent years, Japan has grappled with a rapidly aging population, leading to pressing issues in emergency medical care and an uptick in ambulance services. Our research team, collaborating with the Kobe City Fire Department, aims to address this by devising a predictive model for strategic deployment of medical services. By analyzing extensive EMS data and population demographics, we aim to forecast long-term emergency transport demand accurately, bypassing machine learning algorithms. The results of the forecast revealed regional disparities in future emergency demand, providing indicators for optimal deployment of emergency resources and strategic planning for emergency services.

In pursuit of realizing Society 5.0, this study explores efficient development methods for services tailored to individual user needs. The rapid evolution of digital devices and the diversification of user demographics have led to swiftly changing service needs, necessitating increased personalization. This research focuses on developing technologies that enable service development based on a deep understanding of specific user needs. By utilizing a virtual agent (VA)-based interactive need extraction system developed in previous research, and leveraging the Sharing Up cycling Cases with Context and Evaluation for Efficient Software Development System (SUCCEED System), we aim to automate the extraction of user needs and provide insights to developers. We propose an interactive need extraction method for novel, undeveloped services and a method for obtaining development cases based on these needs, thereby contributing to the efficiency of personalized service development approaches.

The increase in single-person households and the onset of “COVID-19 frailty” due to COVID-19 have become problems. Our research group focuses on technology support and develops virtual agents using speech recognition technology. In this service, we examined how to promote self-care by having the agent listen to the older person and increase conversation opportunities. However, this service has the problem that it cannot increase opportunities for dialogue with family and friends, which leads to mutual support. In this study, we aim to propose and implement a service called “Easy Video Chat Service” that increases the opportunities for conversation with others through a virtual agent listening service for the elderly person at home.

Japan is facing the challenge of an aging society with a significant increase in the number of dementia patients, making it an urgent social issue. Ac-cording to the Ministry of Health, Labour and Welfare, it is estimated that by 2040, the population of elderly individuals aged 65 and above will reach 35.3 % of the total population, with over 7 million of them being dementia patients. Against this backdrop, measures for dementia prevention and support for elderly dementia patients are deemed necessary. Memory impairment is one of the core symptoms of dementia, and current strategies include recording and presenting information by caregivers and using memory aid tools such as notepads for rehabilitation. However, these methods have their limitations, prompting the need for new memory aid techniques that reduce caregiver burden and are sustainable for dementia patients. Therefore, recording daily activities and preserving them as docu-ments is proposed as a method for memory assistance. This study aims to verify the effectiveness of this technique by setting evaluation criteria, conducting self-recordings of daily activities, creating explanatory texts by multiple individuals, and subjectively eval-uating texts generated by recognition AI. Based on the results, the study examines the extent to which the texts generated by recognition AI are useful for memory assistance and provides insights accordingly.

In this study, we describe the design and implementation of a smart service for automated management of hybrid meeting spaces. Hybrid formats, where participants interact both online and offline, are becoming common in modern meeting rooms. To meet this new need, we have developed a remote task execution service using VNC and WebDriver. This service automates the launching of online services and website operations during meeting preparations, greatly reducing the time and effort required by the user. Specifically, the service provides functions to automatically execute multiple remote control tasks, such as launching Zoom and a web site for meeting minutes. In this paper, we confirm the actual operation of the system through a prototype implementation in our laboratory.

The number of elderly people living at home (requiring nursing care or living alone) has been increasing due to the aging of society in recent years. As the elderly stay at home for long periods of time, their connection with the local community becomes less and less, and loneliness becomes a serious problem. In our research group, we are developing a system that allows elderly people to interact with the virtual agent. The purpose of this paper is to develop a method for generating personalized dialogue in order to realize continuous and sentimental dialogue. As the approach, morphological analysis and sentiment analysis are performed on the dialogue content (text) obtained from speech recognition. Then, we present a new method for generating personalized dialogues based on related conversation log summarization. In this way, it is promising to realize smarter personalized dialogues that do not rely on manual creation.

In advanced smart systems, heterogeneous distributed services are integrated dynamically, based on various contexts in cyber/physical worlds. Currently, the logic of such service integration is implemented specifically in each application. Thus, as the way of integration becomes sophisticated, the complexity and development effort of the application become quite expensive. In this research, we propose a service, called Uni-messe (Unified Rule-based Message Delivery Service), which provides context-aware service integration in an application-neutral manner. Based on ECR (Event-Condition-Routing) rules, Uni-messe routes an event message from an application to a designated application based on a specified condition. We first present the architecture of Uni-messe with publish/subscribe messaging platform. We then propose the data model of the ECR rules based on 6W1H (Who/Whom/When/Where/What/Why/How) event model. Finally, we define the condition evaluation and routing behaviors. Using the implementation of Uni-messe, we demonstrate automatic curtain control and daily routine automation in a smart home. Since Uni-messe de-couples the rule-based service integration from individual applications, it allows developers and users to implement and reuse flexible integration efficiently.

Japan is currently facing super-aging society and the assistive technology for self-help and mutual aid of the elderly is becoming urgent. The purpose of this paper is to build a system that can execute various services through dialogue with agents, in order to support elderly people who cannot use Internet services due to lack of access to devices. To achieve the goal, we discuss a framework for executing microservices using dialogue agents. More specifically, the framework consists of the next two essential elements: (1) Managing user information for the various microservices centrally. (2) Configuring the behavior of the agents when executing the services correctly. In the proposed method, we first discuss each element in detail. Then, we demonstrate the effectiveness of the framework by applying it to the actual integration of a dialogue agent and several microservices.

The emerging deep learning technology is a promising means for context recognition with multimedia data. We are interested in using the deep learning with images for context recognition in smart homes. In the home context recognition, the room layout, the environment, and the contexts to be recognized are different from one household to another. Therefore, a unique recognition model is required for every different household. For this, if we take a naive approach that uses the deep learning directly, a huge amount of labeled images are required, which is practically impossible for general households. The goal of this research is to develop an image-based context recognition method that is affordable at home. In the proposed method, we exploit a cognitive API which performs general image recognition, and retrieve the information within the image as text. By using the text as features, we classify the context with ordinal supervised machine learning. Compared with the expensive approach with deep learning, the proposed method uses generic image recognition of the cognitive API, and light-weight machine learning. As a result, the context recognition customized for every household can be achieved with much less effort.