Computer numerical control (CNC) tool-wear prediction (TWPred) is an important issue in the industry. Recently, researchers have demonstrated that deep-learning models (DLMs) are effective in TWPred. However, DLMs are ill-suited to small- and medium-scale manufacturers due to high computational costs. Methods exist to reduce the computational costs of DLMs, but most of them depend on overly-complex pruning processes that are not appropriate for the low-end computers used by the above manufacturers. Therefore, we developed a lightweight DLM and an automated framework for TWPred. The framework is based on two concepts: 1) the DLM was pruned by reducing the number of input data fields so the model itself remains unchanged and 2) we designed a framework that enables the automatic establishment of a lightweight DLM. These two concepts make the overall framework applicable to small- and medium-scale manufacturers. Finally, we used real-world dataset PHM 2010 to verify that the lightweight DLM can achieve almost the same reminding useful life (RUL) accuracy as the DLM (DLM: 95.55% and lightweight DLM: 95.51%) using only 0.88% of DLM parameters, which verifies the low cost and high precision of the proposed model.
27 March 2024 in IEEE Transactions on Artificial Intelligence
The accurate and rapid transfer of complex urban spatio-temporal data is crucial for urban computing tasks such as urban planning and public transportation deployment for smart-city applications. Existing works consider auxiliary data or propose end-to-end models to process complex spatio-temporal information into more complex deep features. However, the latter is incapable of decoupling spatio-temporal knowledge, which means these end-to-end models lack modularity and substitutability. A general modular framework that can automatically capture simple representations of complex spatio-temporal information is required. In this paper, we thus propose a universal framework for the transfer of spatio-temporal knowledge based on a radial basis function (RBF). We termed this approach STRBF-TF. The proposed STRBF-TF generates simple RBF representations of spatio-temporal flow distribution with an RBF transfer block and also leverages a channel attention mechanism. Moreover, we propose two RBF kernel initializers suitable for the source and the target domains, respectively. The framework retains important spatio-temporal knowledge in simple representations for the reconfiguration of spatio-temporal feature distribution for fast and accurate transfer. We conducted cross-domain learning experiments on a large real-world telecom dataset. The results demonstrate the efficiency and accuracy of the proposed approach, as well as its suitability for real-world applications.
Shared transportation, which allows commuters to share vehicles, either through riding in the same vehicle (i.e., ride-sharing) or using the same vehicle at different times (i.e., car-sharing or bike-sharing) has become increasingly popular. Car-sharing and bike-sharing require efficient allocation of vehicle resources to sharing stations. Scholars have used temporal or spatial information to predict the number of users at each station. However, external factors, such as special events or rain, can affect this number. This paper proposes a framework to improve the prediction of shared-transport users based on both temporal and spatial factors as well as the external factors of the surrounding environment of the station, the weather, and relevant online activity. The proposed approach was verified through the application to the real-world case of bicycle-sharing in Taipei, Taiwan.
With vast choices available in modern cities and information explosion on the internet, people face overwhelming decision paralysis when selecting locations for dining, shopping, or entertainment. Point-of-Interest (POI) recommendation systems address this challenge by understanding user context through sequential mobility patterns rather than treating locations as isolated decisions. However, existing approaches face critical limitations: traditional collaborative filtering (CF) and matrix factorization (MF) methods struggle with semantic understanding and cold-start problems by treating POIs as numeric identifiers without contextual meaning, while recent large language model approaches often sacrifice personalization for generalization and inadequately model temporal dynamics between visits, overlooking the realistic time intervals crucial for capturing mobility patterns. We propose GeoChronos-GPT, a spatiotemporal transformer framework that addresses these limitations through three key innovations: hybrid spatiotemporal encoding that captures both semantic relationships and geographic context without sacrificing individual user patterns, explicit temporal feature integration that models realistic mobility dynamics by encoding time intervals between visits, and user-specific POI masking that maintains personalization while leveraging global patterns by constraining predictions to each user’s historical preference space during inference. Our comprehensive experiments on the TSMC-2014 dataset demonstrate that GeoChronos-GPT consistently outperforms existing baselines including traditional CF, MF, and modern sequence-based methods, effectively addressing the semantic understanding, temporal modeling, and personalization challenges that have limited existing POI recommendation approaches.
8 May 2026 in ACM Transactions on Spatial Algorithms and Systems
The selection of key grid cells is critical for enhancing the interpretability and reliability of urban spatial-temporal flow prediction. Manual selection by domain experts can be inconsistent and time-consuming, while recent deep learning approaches that rely on feature visualization often produce unstable results due to sensitivity to input changes. In this work, we propose CRBF-Net, a convolutional radial basis function network that decouples feature representations from prediction outputs by introducing spatially localized basis kernels to prioritize selection stability. Our proposed method is evaluated on real-world telecom mobility data from Taipei, Taiwan. Experimental results demonstrate that CRBF-Net ensures high selection consistency and enhances the transferability of spatial insights across various downstream prediction models, including LSTM, ANN, and KNN. MAPE reductions were observed across all locations, with improvements of up to 8.24% compared to prior visualization-based selection strategies. Additionally, stability analysis via the Jaccard Index confirms that CRBF-Net maintains significantly higher spatial consistency across multiple independent trials. These results highlight the robustness and practical utility of CRBF-Net in selecting meaningful spatial regions, providing a stable and generalizable approach to key grid cells selection in urban flow prediction while contributing to the development of trustworthy and interpretable GeoAI systems.