Wearable accelerometers and gyroscopes capture fine-grained behavioral signatures that can inadvertently reveal user identities, making privacy protection essential for healthcare applications. We present C-AAE, a lightweight compressive anonymizing autoencoder that performs on-device privacy filtering at the sensor edge. The core idea of C-AAE is to integrate two complementary privacy filters: a learned, sensor-specific anonymization module, the Anonymizing AutoEncoder (AAE), and a learning-free, generic anonymization module, Adaptive Differential Pulse-Code Modulation (ADPCM). The AAE locally learns to suppress identity cues while preserving activity-relevant representations, whereas ADPCM provides training-free anonymization through compression, further masking residual identity information and reducing communication cost. Experiments on the MotionSense and PAMAP2 datasets show that C-AAE cuts user re-identification F1 scores by 10–15 percentage points relative to AAE alone, while keeping activity-recognition F1 within 5 percentage points of the unprotected baseline. Implementation on a small-scale edge device (ESP32-WROOM-32) demonstrates real-time performance with markedly lower memory usage, latency, and power consumption. Unlike differential-privacy mechanisms that rely on randomized noise, C-AAE offers a complementary, representation-level approach, enabling practical and resource-efficient on-device anonymization that remains compatible with formal DP frameworks for hybrid deployment on edge healthcare devices.

As digital innovation progresses, individuals are increasingly, often unknowingly, exposed to digital temptations and deceptive tactics. These risks, now magnified by foundation AI models, call for protective measures beyond what current digital health technologies can offer. This article introduces Health Behavioral Security , a novel AIoT-driven framework that positions trusted AIoT devices as a distributed “System 0” layer—an always-on, pre-conscious filter that can sense, interpret, and modulate persuasive cues in real time. Guided by five behavior-centric constructs—Assets, Threats, Vulnerabilities, Risks, and Countermeasures—the framework could establish a secure behavioral ecosystem in which adaptive sensing and nudging loops help preserve autonomy while promoting well-being. We present this vision, outline key research challenges, and discuss future directions for advancing AIoT-based behavioral security in today’s increasingly digital society.

This paper presents ZEL+, a wearable lifelogging system designed to operate with net-zero energy consumption by leveraging multiple energy harvesting technologies for continuous context sensing. Self-powered wearable devices often encounter difficulties in environments with inconsistent or low-intensity ambient energy, particularly in indoor settings. To address this challenge, ZEL+ incorporates three key design features. First, it employs a power-switching mechanism based on dual comparators and a capacitor to manage surplus energy and support operation under varying lighting conditions. Second, the system integrates heterogeneous energy harvesters not only as power sources but also as sensing elements. Specifically, a dye-sensitized solar cell provides stable responses under low-light indoor environments, while an amorphous solar cell exhibits sensitivity to changes in ambient illumination; together with a piezoelectric element capturing motion-induced signals, these components contribute complementary cues for location and activity recognition. Third, a Spatial Consistency-Based Correction (SCC) algorithm is applied as a post-processing step to mitigate transient recognition errors and improve the coherence of inferred lifelogs. The system is implemented as a 192 g nametag-shaped wearable device and evaluated in a real-world office environment with 11 participants. Under a person-dependent setting, ZEL+ achieved an accuracy of 96.62% for 8-location place recognition and 97.09% for static/dynamic activity recognition, while maintaining robust performance on more fine-grained tasks. In terms of energy sustainability, the device sustained autonomous operation using harvested energy alone for approximately 93.97% of a standard 8-hour office workday. These results indicate that ZEL+ provides a practical and energy-sustainable solution for continuous lifelogging in indoor mobile computing environments.

Decentralized Machine Learning (DML) enhances machine learning by improving scalability, adaptability, and privacy. Federated Learning (FL), a DML approach, boosts privacy by transmitting model gradients from local training to a central server instead of raw data. However, FL encounters challenges like potential privacy risk through gradient inference and communication bottlenecks. Several studies implement Differential Privacy (DP) to tackle these issues by adding random noise to gradients before aggregation. This approach safeguards sensitive information but can decrease model accuracy due to the introduced noise. Adaptive clipping helps by dynamically adjusting clipping bounds based on data, minimizing noise, and preserving data utility. However, in a decentralized system, noise application may be inconsistent across nodes, leading to learning inefficiencies. Our solution involves clustering, enabling local node collaboration and model aggregation to reduce DP noise impact and decentralize aggregation, minimizing bottlenecks caused by the central server. We also use data sharding to balance datasets across nodes, ensuring each processes representative data portions. This improves the signal-to-noise ratio, addresses data imbalance, and enhances accuracy. Experiments with MNIST and Fashion-MNIST datasets demonstrate that our method achieves better accuracy than conventional DP-SGD algorithms.

Since office work has become large-scale and diversified in companies or organizations, work engagement and efficiency have been always an important index of a team's or group's evaluation because it is directly connected to their outcomes. In order to identify the group work context, we first need to recognize for what and how long the individual members are spending their time at their desks, but without privacy concerns and underestimation of their actual work. In this paper, we propose and evaluate the base system of personal desk activity recognition by using a low-cost compact WiFi node and its WiFi channel state information (CSI), which can lead to a lightweight group work context identification system. As a result, we achieved 94.2% desk activity recognition accuracy using the on-desk receiver, in recognizing five different classes.

We present a secure decentralized learning framework suitable for resource-constrained devices within a cluster environment. Our approach focuses on enhancing privacy preservation during model aggregation by utilizing Differential Privacy. This technique adds random noise to gradients obtained from local training on edge devices before sending them for aggregation. This noise addition ensures that sensitive information within the gradients remains distorted, thus safeguarding user privacy. We showcase the implementation of our system on a cluster system employing Raspberry Pi 4 Model B devices, illustrating its feasibility and effectiveness in real-world scenarios. Through this demonstration, we highlight the practical applicability of our system in enabling secure decentralized learning within resource-constrained environments.

Indoor activity recognition using WiFi sensing is expected to have a wide range of applications, such as monitoring the elderly and home security. The state of radio wave propagation is called Channel State Information (CSI) and can be obtained using specific devices. By collecting CSI and applying machine learning, it is possible to recognize activities. However, CSI is sensitive to changes in the environment, so whenever the arrangement of furniture or the layout of the room changes, it is necessary to re-collect sample data and retrain the model. Retraining a model requires annotation work, which is costly in terms of time and effort. To address this issue, this paper proposes an annotation system that uses backscatter tags to reduce the cost of data collection and model training. In this system, a backscatter tag that generates a frequency shift depending on its angle is attached to a person during data collection, and activity recognition is performed by detecting the presence of the frequency shift. The backscatter tag-based recognition results are then used as pseudo-ground truth for model update.

Federated learning (FL) has gained widespread attention as a distributed machine learning (ML) technique that offers data protection when training on local devices. Unlike conventional centralized training in traditional ML, FL incorporates privacy and security measures as it does not share raw data between the client and server, thereby safeguarding potentially sensitive information. However, there are still vulnerabilities in the FL field, and commonly used approaches, such as encryption and blockchain technologies, often result in significant computational and communication costs, making them impractical for devices with restricted resources. To tackle this challenge, we present a privacy-preserving FL system specifically designed for resource-constrained devices, leveraging compressive sensing and differential privacy (DP) techniques. We implemented the weight-pruning-based compressive sensing method with an adaptive compression ratio based on resource availability. In addition, we employ DP to introduce noise to the gradient before sending it to a central server for aggregation, thereby protecting the gradient’s privacy. Evaluation results demonstrate that our proposed method achieves slightly better accuracy when compared to state-of-the-art methods like DP-federated averaging, DP-FedOpt, and adaptive Gaussian clipping-DP (AGC-DP) for the MNIST, Fashion-MNIST, and Human Activity Recognition data sets. Furthermore, our approach achieves this higher accuracy with a lower total communication cost and training time than the current state-of-the-art methods. Moreover, we comprehensively evaluate our method’s resilience against poisoning attacks, revealing its better resistance than existing state-of-the-art approaches.

In this paper, we propose ToonMeet, a hybrid frame-work for high-resolution and style-controllable online meeting toonification that ensures real-time operation speed. ToonMeet applies video frame interpolation to traditional portrait toonification pipelines, allowing for the synthesis of intermediate frames between adjacent toonified keyframes, significantly accelerating the overall process and saving computational resources. However, this approach brings a new problem, where prevailing flow-based video frame interpolation methods tend to cause more ghost and blur artifacts in toonified scenes compared to non-toonified scenes, especially when fast-moving objects exist. We study this previously undiscussed problem and explore its causes. To address this, we introduce a new dataset called TM3B (Toonified Multi-modal Meeting Behaviors), offering high-resolution and cross-platform multi-modal stylized meeting data of Japanese youth in various scenarios. Then, we fine-tune ToonMeet on these tailored data and the resulting model presents improved optical flow estimation ability on toonified videos. Extensive experiments demonstrate that ToonMeet can achieve great spatiotemporal performance and perform high-quality toonification of online meetings with real-time operation speed.

Federated learning has gained popularity as a distributed machine learning approach that provides security and privacy for data trained on local devices. However, vulnerabilities still exist in this approach, and common solutions such as encryption and blockchain techniques often suffer from high computation and communication costs, making them impractical for resource-constrained devices. To solve this problem, we propose a privacy-preserving federated learning system that leverages compressive sensing and differential privacy, specifically designed for devices with limited computational resources. In this paper, we demonstrate the capabilities of our proposed system in resource-limited environments. We outline the features, infrastructure, and algorithm of our proposed system, and simulate its performance using image datasets on a Raspberry Pi 4 and an Android smartphone in a cloud environment. Our approach offers a practical solution for secure and privacy-preserving federated learning in resource-constrained scenarios, with potential applications in various domains such as healthcare, IoT, and edge computing.

Federated learning provides techniques for training algorithms using mobile or decentralized devices, in contrast to traditional machine learning in which algorithm training is performed on centralized devices. In addition, federated learning provides privacy and security features, as the client and server do not share raw data, which may contain confidential information. A number of studies have shown, however, that using federated learning alone is not enough to protect data privacy in certain situations. To overcome this problem, differential privacy is proposed, which is a technique in which artificial noise is added to the raw data. By implementing this method, a high level of privacy protection can be obtained, however this added noise also reduces model accuracy. To address this issue, this paper proposes a new approach to implement differential privacy in federated learning using adaptive Gaussian clipping. We implemented the method by tightening the privacy budget, and introducing dynamic sampling probability, adaptive clipping based on hyperparameters, and a new privacy loss calculation. Our method’s main objective is to adaptively change the amount of noise given to the model, thereby maximizing the model’s accuracy performance, while maintaining privacy protection levels. Evaluation results show that our proposed method presents slightly better accuracy when compared to other existing differential privacy variants such as RDP, DP-SGD, and ZcDP, for both balanced (i.i.d.) and unbalanced datasets (non-i.i.d.), for a lower total communication cost than some variants.

We present ZEL, the first net-zero-energy lifelogging system that allows office workers to collect semi-permanent records of when, where, and what activities they perform on company premises. ZEL achieves high accuracy lifelogging by using heterogeneous energy harvesters with different characteristics. The system is based on a 192-gram nametag-shaped wearable device worn by each employee that is equipped with two comparators to enable seamless switching between system states, thereby minimizing the battery usage and enabling net-zero-energy, semi-permanent data collection. To demonstrate the effectiveness of our system, we conducted data collection experiments with 11 participants in a practical environment and found that the person-dependent (PD) model achieves an 8-place recognition accuracy level of 87.2% (weighted F-measure) and a static/dynamic activities recognition accuracy level of 93.1% (weighted F-measure). Additional testing confirmed the practical long-term operability of the system and showed it could achieve a zero-energy operation rate of 99.6% i.e., net-zero-energy operation.

Maintaining hand hygiene is the one of the most effective way to prevent the spread of germs during a pandemic. This paper focuses on encouraging people to use a hand sanitizer more frequently by applying the nudge theory to improve hand hygiene behavior in private organizations. We propose a system that recognizes hand hygiene behavior using face recognition and detects hand sanitizer use. The system responds to the user's personal hand hygiene behavior with animation of a virtual bonsai as an interactive agent. To preserve user privacy, we implemented the system on an edge device and conducted experiments for 4 case studies in 2 real-world organizations. The results showed that the system improved the hand hygiene behavior of people in a private organization.

In response to the Coronavirus 2019 (COVID-19) pandemic, the World Health Organization (WHO) has published preventive measures such as performing hand hygiene frequently, wearing a medical mask, trying to avoid touching face and so on. This paper presents a nudge-based system to promote hand hygiene in a private organization. The proposed system consists of a hand sanitizer station equipped with a magnetic sensor to sense user presses. We conducted 4 case studies to compare the effects of nudging on the frequency of hand sanitizer use: no nudging, traditional nudging, non-personalized nudging, and personalized nudging. The results reveal that using nudge-based methods offer a significant increase in the frequency hand sanitizer use.

Recently smart homes equipped with many sensors and IoT devices are widespread. However, when smart home users receive smart home services like elderly monitoring, they need to upload their privacy sensitive data to potentially untrusted cloud servers where the service quality (user's benefit) depends on the amount/frequency of the uploaded data. In this paper, aiming to minimize the risk of privacy leakage and maximize users' benefit obtained through services, we propose a novel privacy-aware data management method that works on a smart-home system composed of smart homes with sensors, edge computing servers, and a cloud server. We formulate a combinatorial optimization problem which determines the best choice of data type (raw or activity label recognized at the edge) and upload frequency in each time slot taking into account the constraints of edge server resources and users' budgets as well as the k-anonymity of activities and users' preferences. Since the target problem is NP-hard, we propose a heuristic algorithm to derive semi-optimal solutions by determining choices with better objective function values in a greedy manner. Through experiments using smart-home open dataset, we confirmed that the proposed method outperforms the conventional methods using only a cloud server.

A wearable based long-term lifelogging system is desirable for the purpose of reviewing and improving users lifestyle habits. Energy harvesting (EH) is a promising means for realizing sustainable lifelogging. However, present EH technologies suffer from instability of the generated electricity caused by changes of environment, e.g., the output of a solar cell varies based on its material, light intensity, and light wavelength. In this paper, we leverage this instability of EH technologies for other purposes, in addition to its use as an energy source. Specifically, we propose to determine the variation of generated electricity as a sensor for recognizing "places" where the user visits, which is important information in the lifelogging system. First, we investigate the amount of generated electricity of selected energy harvesting elements in various environments. Second, we design a system called EHAAS (Energy Harvesters As A Sensor) where energy harvesting elements are used as a sensor. With EHAAS, we propose a place recognition method based on machine-learning and implement a prototype wearable system. Our prototype evaluation confirms that EHAAS achieves a place recognition accuracy of 88.5% F-value for nine different indoor and outdoor places. This result is better than the results of existing sensors (3-axis accelerometer and brightness). We also clarify that only two types of solar cells are required for recognizing a place with 86.2% accuracy.

Information and communication technologies (ICTs) has enabled growth in developed countries and urban cities through improvements in communication systems, devices and applications. In rural areas, especially in developing countries, ICT penetration is not as high, often due to lack of available infrastructure and funding. With the increasing availability of Internet-of-Things (IoT) devices, low-cost large-scale deployments have become possible even in rural areas. We design, develop and implement, Near Cloud, a cloud-less platform that allows users and IoT devices to communicate and share information. This is built on top of a wireless mesh network (WMN) of low-cost, low-power IoT devices and deployed in areas where there is little to no Internet connectivity. To inject ICT and help bridge the digital divide in rural areas, Near Cloud provides functionalities such as web servers on nodes, accessibility to all users via Wi-Fi, and various data processing including image processing and machine learning. We will show applicability of Near Cloud in improving rural education, health care facilities, disaster response and agriculture.

In an era where billions of IoT devices are deployed, edge/fog computing paradigms are attracting attention for their ability to reduce processing delays and mitigate waste of communication resources. However, since the computing system assumed by edge/fog paradigms have heterogeneity (in terms of the computing power of devices, network performance between devices, device density, etc.), provisioning computational resources according to computational demand becomes a challenging constrained optimization problem. In this paper, we propose in-situ resource provisioning method consisting of insitu resource area selection with adaptive scale out and in-situ task scheduling based on tabu search algorithm. We conducted a simulation study in a target regional area where 2,000 IoT devices and 10 IoT services are deployed to evaluate the effectiveness of the proposed algorithm. The simulation results show that our proposed algorithm can obtain higher user QoS compared to conventional resource provisioning algorithms.

EdgeComputing and Fog Computing are new paradigms where data processing is executed in or on the edge of networks to mitigate cloud server load. However, EdgeComputing and Fog Computing still need powerful servers on the edge of networks which impose additional costs for deployments. We proposed a platform called IFoT (Information Flow of Things) that efficiently performs distributed processing as well as distribution and analysis of data streams near their sources based on "Process On Our Own (PO3)" concept. In IFoT, processing of tasks for cloud servers is delegated to an ad-hoc distributed system consisting of proximity IoT devices for distributed real-time stream processing. In this demonstration, we show a face recognition system for person tracking developed on top of IFoT middleware which locally processes video streams in real-time and in a distributed manner by using computational resources of IoT devices.