Yugo Nakamura

Taste is not shaped by the tongue alone: smell and vision also influence how a beverage is experienced. We present Kaolid+, a lid-type olfactory interface that routes aroma either toward the nose or through a straw, while illuminating the drink with RGB light. This form factor allowed us to compare orthonasal and retronasal aroma delivery under a closely matched setup, and to compare subjective taste ratings across no-light, aroma-congruent orange-light, and aroma-incongruent blue-light conditions under retronasal orange aroma delivery. In a within-subjects design with unflavored carbonated water and orange aroma, the retronasal condition received higher Perceived Sweetness and Flavor Change ratings than the orthonasal condition; overall acceptability did not differ significantly between the two methods. Under retronasal delivery, the orange-light condition was associated with higher Perceived Sweetness ratings than the no-light condition, and the orange-light condition was rated higher than the blue-light condition for overall acceptability, sweetness, and naturalness. Kaolid+ received a mean SUS score of 77.63, indicating good usability. The results position Kaolid+ as a smart beverage interface for studying lid-integrated olfactory actuation and visual cueing during everyday drinking, and provide initial evidence that these device conditions may be associated with changes in sweetness-related ratings within the tested orange-aroma carbonated-water scenario.

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.

Egocentric video captured by wearable cameras offers rich contextual information for recognizing human activities in daily life. However, such video often includes sensitive personal details that must be protected from external threats. This creates a fundamental trade-off between preserving data utility and ensuring privacy, particularly in scenarios where continuous activity monitoring is required. In this study, we explore the concept of using EdgeVLM—a vision-language model designed to run entirely on edge devices—as a privacy filter for wearable camera data. We investigate the impact of caption granularity and demonstrate that our method locally transforms egocentric video into semantically rich textual image captions, enabling activity detection without transmitting raw visual content to the cloud. This edge-based processing preserves contextual cues while minimizing privacy risks through data minimization. To evaluate this approach, we conducted a quantitative user study (N=88) that found EdgeVLM-generated captions notably decreased participants’ privacy concerns compared to raw, blurred, or cartoonized images. Critically, from a bystander’s perspective, the proposed method demonstrated privacy protection levels statistically comparable to canny edge detection. Additionally, when combined with accelerometer data, the caption-based method achieved a 77.2% accuracy in recognizing desk activities such as typing, mousing, swiping, drinking, and writing—effectively replacing pixel-level visual information with text while maintaining performance comparable to models using unfiltered visuals. These findings indicate that EdgeVLM-based image captioning is a promising privacy-conscious solution for wearable camera applications, facilitating continuous activity recognition while protecting user privacy at the edge.

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.

Human activity recognition (HAR) using 1-D sensor signals is a widely studied domain and plays an important role in intelligent systems. Traditional research applied a sliding window to extract fixed-size frames from the original signals and ran a classification process on them to recognize activities. However, sliding-window-based methods suffer from several problems, such as the multi-class windows problem, where all sample points share the same label in a frame even if there is more than one activity in it. The sliding window also makes it challenging to decide the best window length for all activities and causes the computational cost problem due to the redundant calculation of two consecutive frames. To solve the problems of sliding windows, in this paper, we design a novel sliding-window-free method that can recognize all activities from signal segments in a highly efficient way. As a solution, we propose grid transition labeling, which is derived from the object detection tasks in 2-D images, and we introduce new designs to deal with 1-D signal-based HAR tasks. Grid transition labeling regards the transition between two activities as an object. To detect transitions, grid transition labeling divides a segment into several grids and labels whether there is a transition in each grid, as well as the classes around transitions, therefore all sample points of the segment can be well labeled without the need of labeling each of them. To evaluate our method, we conducted experiments on two datasets, including UCI HAPT and WISDM. The results on the two datasets show accuracies of 0.930 and 0.939 respectively, outperforming two sliding-window-free methods, FCN and U-Net. Our method also has a high computational efficiency that requires much less computational cost for inference compared to both the sliding-window-based methods and the sliding-window-free methods, making it more suitable for applications on edge devices such as smartwatches and smartphones.

This article envisions HabiComp—a paradigm shift in ubiquitous computing that redefines technology not just as a tool for seamless interaction, but as an ethical architect of habitual behavior. This project has been exploring nonverbal nudging through System 0, an ambient, context-aware cognitive preprocessing layer that supports human cognitive Systems 1 and 2 by recalibrating habitual actions while preserving autonomy. Two case studies illustrate its subtle yet profound impact: one reconfigures eating behavior through environmental cues, to foster mindful consumption, and the other reshapes digital reward structures to ethically disengage users from compulsive behaviors. Building on these insights, this article outlines key challenges and future directions, positioning HabiComp as a guiding framework for the ethical redesign of habits within ubiquitous computing.