This manuscript presents GazeBase, a large-scale longitudinal dataset containing 12,334 monocular eye-movement recordings captured from 322 college-aged participants. Participants completed a battery of seven tasks in two contiguous sessions during each round of recording, including a – (1) fixation task, (2) horizontal saccade task, (3) random oblique saccade task, (4) reading task, (5/6) free viewing of cinematic video task, and (7) gaze-driven gaming task. Nine rounds of recording were conducted over a 37 month period, with participants in each subsequent round recruited exclusively from prior rounds. All data was collected using an EyeLink 1000 eye tracker at a 1,000 Hz sampling rate, with a calibration and validation protocol performed before each task to ensure data quality. Due to its large number of participants and longitudinal nature, GazeBase is well suited for exploring research hypotheses in eye movement biometrics, along with other applications applying machine learning to eye movement signal analysis. Classification labels produced by the instrument’s real-time parser are provided for a subset of GazeBase, along with pupil area.

Many developers of biometric systems start with modest samples before general deployment. However, they are interested in how their systems will work with much larger samples. To assist them, we evaluated the effect of gallery size on biometric performance. Identification rates describe the performance of biometric identification, whereas ROC-based measures describe the performance of biometric authentication (verification). Therefore, we examined how increases in gallery size affected identification rates (i.e., Rank-1 Identification Rate, or Rank-1 IR) and ROC-based measures such as equal error rate (EER). We studied these phenomena with synthetic data as well as real data from a face recognition study. It is well known that the Rank-1 IR declines with increasing gallery size, and that the relationship is linear against log(gallery size). We have confirmed this with synthetic and real data. We have shown that this decline can be counteracted with the inclusion of additional information (features) for larger gallery sizes. We have also described the curves which can be used to predict how much additional information would be required to stabilize the Rank-1 IR as a function of gallery size. These equations are also linear in log(gallery size). We have also shown that the entire ROC-curve was not systematically affected by gallery size, and so ROC-based scalar performance metrics such as EER are also stable across gallery size. Unsurprisingly, as additional uncorrelated features are added to the model, EER decreases. We were interested in determining the impact of adding more features on the median, spread and shape of similarity score distributions. We present evidence that these decreases in EER are driven primarily by decreases in the spread of the impostor similarity score distribution.

Various sensors have been proposed to address the negative health ramifications of inadequate fluid consumption. Amongst these solutions, motion-based sensors estimate fluid intake using the characteristics of drinking kinematics. This sensing approach is complicated due to the mutual influence of both the drink volume and the current fill level on the resulting motion pattern, along with differences in biomechanics across individuals. While motion-based strategies are a promising approach due to the proliferation of inertial sensors, previous studies have been characterized by limited accuracy and substantial variability in performance across subjects. This research seeks to address these limitations for a container-attachable triaxial accelerometer sensor. Drink volume is computed using support vector machine regression models with hand-engineered features describing the container’s estimated inclination. Results are presented for a large-scale data collection consisting of 1908 drinks consumed from a refillable bottle by 84 individuals. Per-drink mean absolute percentage error is reduced by 11.05% versus previous state-of-the-art results for a single wrist-wearable inertial measurement unit (IMU) sensor assessed using a similar experimental protocol. Estimates of aggregate consumption are also improved versus previously reported results for an attachable sensor architecture. An alternative tracking approach using the fill level from which a drink is consumed is also explored herein. Fill level regression models are shown to exhibit improved accuracy and reduced inter-subject variability versus volume estimators. A technique for segmenting the entire drink motion sequence into transport and sip phases is also assessed, along with a multi-target framework for addressing the known interdependence of volume and fill level on the resulting drink motion signature.

A lightweight on-device liquid consumption estimation system involving an energy-aware machine learning algorithm is developed in this work. This system consists of two separate on-device neural network models that carry out liquid consumption estimation with the result of two tasks: the detection of sip from gestures with which the bottle is handled by its user and the detection of first sips after a bottle refill. This predictive volume estimation framework incorporates a self-correction mechanism that can minimize the error after each bottle fill-up cycle, which makes the system robust to errors from the sip classification module. In this paper, a detailed characterization of sip detection is performed to understand the accuracy-complexity tradeoffs by developing and implementing a variety of different ML models with varying complexities. The maximum energy consumed by the entire framework is around 119 $\mathrm{mJ}$ during a maximum computation time of 300 $\mathsf{\mu }\mathrm{s}$. The energy consumption and computation times of the proposed framework is suitable for implementation in low-power embedded hardware that can be incorporated in consumer grade water bottles.

This paper introduces a new approach to incentivise the vehicles in connected vehicle (CV) networks based on the reputation measures along with a combination of graph neural network and reinforcement learning (GNN-RL). The proposed method enables vehicles to create reputation estimates of their nearby vehicles by analyzing broadcasted kinematic data and onboard sensor estimates, as well as the network connectivity topology. This data is then utilized to create a graphical representation of reputation distribution. A centralized RL agent is used for providing reward signals to each vehicle based on a Laplacian matrix, which encourages the vehicles to make more accurate reputation estimates. The proposed algorithm is based on a GNN-RL algorithm previously used for coordinated navigation, which has been adapted to the cybersecurity domain in this paper. The simulation results show that the model was effective in giving dynamic rewards to vehicles based on their reputation scores. Further, Laplace matrices helped in analyzing the connectivity and behavior of CVs in the network.

Refillable smart bottles providing automated water consumption tracking are a valuable tool for aiding in the achievement of hydration goals. We have previously proposed an attachable accelerometer sensor capable of providing retrofittable consumption tracking functionality to any bottle. This paper expands upon our prior work by demonstrating inclination tracking using a six degree-of-freedom inertial measurement unit sensor. A simplified approach for fusing information from the accelerometer and gyroscope sensors is proposed herein. Estimates are verified using video-based motion capture performed in Blender. A performance improvement of over 25% is achieved through utilization of the proposed fusion approach versus estimates formulated using the accelerometer only.