The increasing adoption of advanced three-dimensional (3D) scanning technologies has made large-scale point clouds containing millions of 3D measurement points standard in applications like manufacturing. However, processing immense amounts of 3D data imposes significant computational loads, often resulting in discarded critical information and suboptimal outcomes for downstream tasks. This paper introduces Point-ITR, a task-oriented sampling method tailored for regression tasks, which selectively retains the most informative points within large-scale point clouds. Specifically, we propose a gradient-based importance sampling framework for intra-sample selection (selecting points within a 3D point cloud) and a feature-based weighting scheme for inter-sample selection (selecting among different 3D point cloud sub-samples). Additionally, we introduce an iterative random sampling (ItrRS) module for preprocessing and an Offset Residual Block that utilizes a reference design model to learn structural features and accelerate both training and testing, which allows a simple fully connected network to process large-scale point clouds. Our approach improves prediction accuracy across downstream tasks while ensuring that the rich details captured are fully utilized for interpretation, offering a more effective and efficient solution. We validate our methodology through simulation studies and real-world case applications in additive manufacturing, demonstrating its robustness and practical applicability.

In this work, we propose a novel activation mechanism called LayerAct for CNNs. This approach is motivated by our theoretical and experimental analyses, which demonstrate that Layer Normalization (LN) can mitigate a limitation of existing activation functions regarding noise robustness. However, LN is known to be disadvantageous in CNNs due to its tendency to make activation outputs homogeneous. The proposed method is designed to be more robust than existing activation functions by reducing the upper bound of influence caused by input shifts without inheriting LN's limitation. We provide analyses and experiments showing that LayerAct functions exhibit superior robustness compared to ElementAct functions. Experimental results on three clean and noisy benchmark datasets for image classification tasks indicate that LayerAct functions outperform other activation functions in handling noisy datasets while achieving superior performance on clean datasets in most cases.

Forecasting complex time series is an important yet challenging problem that involves various industrial applications. Recently, masked time-series modeling has been proposed to effectively model temporal dependencies for forecasting by reconstructing masked segments from unmasked ones. However, since the semantic information in time series is involved in intricate temporal variations generated by multiple time series components, simply masking a raw time series ignores the inherent semantic structure, which may cause MTM to learn spurious temporal patterns present in the raw data. To capture distinct temporal semantics, we show that masked modeling techniques should address entangled patterns through a decomposition approach. Specifically, we propose ST-MTM, a masked time-series modeling framework with seasonal-trend decomposition, which includes a novel masking method for the seasonal-trend components that incorporates different temporal variations from each component. ST-MTM uses a period masking strategy for seasonal components to produce multiple masked seasonal series based on inherent multi-periodicity and a sub-series masking strategy for trend components to mask temporal regions that share similar variations. The proposed masking method presents an effective pre-training task for learning intricate temporal variations and dependencies. Additionally, ST-MTM introduces a contrastive learning task to support masked modeling by enhancing contextual consistency among multiple masked seasonal representations. Experimental results show that our proposed ST-MTM achieves consistently superior forecasting performance compared to existing masked modeling, contrastive learning, and supervised forecasting methods.

Abstract The out-of-home (OOH) advertising market has been operated exclusively following the know-how of salespeople. Thus, it is difficult to make scientific decisions and systematically provide various options to advertisers. In this regard, this study develops an OOH advertising recommendation system by analyzing past OOH history data. The OOH advertising allocation problem has the characteristics that the training data are implicit feedback, and only one advertisement can be posted per offline billboard. This study proposes a recommendation system suitable for OOH history data using negative sampling and Deep Interest Network. The proposed recommendation system showed a higher performance than excisting models used for comparison purposes, and the findings of this study present implications for solving similar recommendation problems.

Manufacturers have implemented continuous multistage manufacturing processes (MMPs) for their efficiency and flexibility, especially in high-volume production of liquid products. While significant attention has been given to the development of data-driven soft sensors in manufacturing fields, research explicitly addressing continuous MMPs of liquid products is very scarce due to the following challenges: obtaining intermediate output labels and determining lead-time between stages. To overcome these challenges, we introduce Multistage Net, a novel machine learning model designed for continuous MMPs of liquid products. In Multistage Net, several interstage blocks are organized in a hierarchical structure within a multistage module. The interstage block is proposed to capture the sequential dependency between the previous and current stages and concurrently explores the lead-time relationship. From the interconnected interstage blocks, the multistage module can learn the sequential nature of MMPs across all stages even in the absence of intermediate output labels. Through validation experiments on two real-world datasets, it is shown that Multistage Net demonstrates superior prediction performance compared to baseline models. Moreover, further analysis reveals that the prediction performance of Multistage Net is not significantly impacted by the nonexistence of lead-time labels.

With the current advancements in mobile and sensing technologies used to collect real-time data in offline stores, retailers and wholesalers have attempted to develop recommender systems to enhance sales and customer experience. However, existing studies on recommender systems have primarily focused on e-commerce platforms and other online services. They did not consider the unique features of indoor shopping in real stores such as the physical environments and objects, which significantly affect the movement and purchase behaviors of customers, thereby representing the "spatiotemporal contexts" that are critical to identifying recommendable items. In this study, we propose a gamification approach wherein a real store is emulated in a pixel world and a recurrent convolutional network is trained to learn the spatiotemporal representation of offline shopping. The superiority and advantages of our method over existing sequential recommender systems are demonstrated through a real-world application in a hypermarket. We believe that our work can significantly contribute to promoting the practice of providing recommendations in offline stores and services.

The health index measures a person’s overall health status which provides useful information for people to manage their health, so developing a precise and relevant health index is urgent. Currently, many researchers have studied the biological age (BA) estimation, one of the beneficial health indices, by applying machine learning and deep learning techniques to health data. However, most of them have focused on the chronological age prediction or basic latent feature extraction methods. In this paper, we present a new algorithm to estimate BA, called Risk Score-Embedded Autoencoder-based BA (RSAE-BA). RSAE-BA can provide an accurate health index by using deep representation learning with an individual’s health risk. We first proposed a notion of risk score (RS) calculation to monitor a person’s health risk. Then we extracted latent features by using an autoencoder embedding the RS, and used them to generate BA. To evaluate RSAE-BA, we presented a new BA validation method using the RS, which is applicable to both unlabeled and labeled data. We compared the results of RSAE-BA with existing methods, and demonstrated the accuracy of RSAE-BA and its applicability to predict disease incidence. We believe that RSAE-BA will be a useful alternative method to measure a person’s health.

Aggressive driving behavior (ADB) is a major cause of traffic accidents. As ADB is controllable, ADB-based driving risk assessment is an effective method for drivers and transportation companies to ensure driving safety. Conventionally, the relationships between ADBs and accident-related records are analyzed when assessing driving risk. However, such records typically overlook driver responsibility for driving risks and depend considerably on the person producing the data (e.g., police officers or insurance managers). Foremost, conventional approaches do not consider non-accident situations that comprise most driving scenarios. Thus, we propose a novel driving risk assessment method that uses only ADB data. In this method, interpretable latent risk factors are extracted from ADB data via sparse non-negative matrix factorization (NMF), and then the driving risk score is computed on a scale of 0–100. The proposed method was validated by adopting a real-world application to assess the driving risk of bus drivers in South Korea and by conducting an evaluation performed by transportation experts in conjunction with the Korea Transportation Safety Authority. Results revealed that the proposed method can discriminate between high- and low-risk driving, thus providing clear guidelines to improve driving. Then, the proposed driving risk score assessment method using NMF was compared with existing machine learning-based risk assessment methods. The proposed method outperformed the conventional methods in terms of driving risk discrimination and interpretability. This study can provide risk assessment guidelines based on driving behavior records and contribute to the application of machine learning in transportation safety management.

Diet planning is a basic and regular human activity. Previous studies have considered diet planning a combinatorial optimization problem to generate solutions that satisfy a diet's nutritional requirements. However, this approach does not consider the composition of diets, which is critical for diet recipients' to accept and enjoy menus with high nutritional quality. Without this consideration, feasible solutions for diet planning could not be provided in practice. This suggests the necessity of diet planning with machine learning, which extracts implicit composition patterns from real diet data and applies these patterns when generating diets. This work is original research that defines diet planning as a machine learning problem; we describe diets as sequence data and solve a controllable sequence generation problem. Specifically, we develop the Teacher-forced REINFORCE algorithm to connect neural machine translation and reinforcement learning for composition compliance with nutrition enhancement in diet generation. Through a real-world application to diet planning for children, we validated the superiority of our work over the traditional combinatorial optimization and modern machine learning approaches, as well as human (i.e., professional dietitians) performance. In addition, we construct and open the databases of menus and diets to motivate and promote further research and development of diet planning with machine learning. We believe this work with data science will contribute to solving economic and social problems associated with diet planning.