Differential-Algebraic Equations (DAEs) play a central role in modeling complex physical systems such as power grids and chemical processes. However, current machine learning approaches for DAE fitting, mainly based on numerical-solver-inspired architectures and polynomial activation functions, suffer from limited generalization and gradient instability. This is primarily due to reliance on fixed coefficient tables in model architectures and the unsuitability of most polynomials as activation functions. This paper argues that integrating NeuralODE—a more flexible and expressive architecture—with a newly designed activation function offers a more effective solution, because NeuralODEs naturally encode the differential properties of physical systems, and when paired with a properly chosen activation function, they can accelerate convergence and ensure the uniqueness of solutions. Specifically, we propose a learnable hybrid combination of a Lipschitz-continuous activation and a DAE-specific basis function within the NeuralODE framework using convex combination. This design alleviates both generalization and stability issues. To further verify its effectiveness, we introduce a novel strategy for observing gradient flow during model training. Our method achieves excellent performance across five distinct DAE tasks, reducing average MSE by several orders of magnitude. It also improves the average neuron activation ratio by 15.5 % and maintains stable gradient flow, suggesting that future research can build upon this direction by emphasizing gradient flow stability in PINN-based frameworks.

Segment Anything Model (SAM) has achieved impressive segmentation performance in an open-world setting. However, SAM relies heavily on high-quality input images and usually struggles in low-light conditions. This is mainly caused by the pre-training dataset, SA-1B, in which low-light samples constitute a relatively small fraction of the data. This lack of presence leads to a noticeable weakness when SAM is applied in real-world dark environments. With the motivation of improving SAM's performance under low-light conditions while retaining its strong zero-shot capability, this work proposes an alignment stage between the pre-training stage and testing stage. Unlike existing low-light studies that mainly focus on task-specific and close-set settings, our work further emphasizes pursuing the segmentation ability under low-light conditions for open-world models. To this end, we construct DarkSeg58K, a realistic and diverse dataset, which serves as the alignment dataset to support this stage. We further introduce Lighted-SAM as the lightweight repair strategy to fix SAM's performance in low-light conditions. Different from existing methods focusing on introducing spectral adapters into the model design and training this model end-to-end, Lighted-SAM introduces the Spectral Information Resonance (SIR) mechanism to harmoniously integrate the spectral enhancement module into SAM, which is usually kept frozen due to its large-scale parameters. Based on our lightweight repairing strategy, Lighted-SAM can improve SAM's ability in low-light conditions while preserving its zero-shot ability. Experiments on different benchmarks validate the superiority of our approach. Code is available at: https://github.com/Jaaaahan/LightedSAM.

Statistical and machine learning methods have been proposed to predict hard drive failure based on SMART attributes, and many achieve good performance. However, these models do not give a good indication as to when a drive will fail, only predicting that it will fail. To this end, we propose a new notion of a drive's health degree based on the remaining working time of hard drive before actual failure occurs. An ensemble learning method is implemented to predict these health degrees: four popular individual classifiers are individually trained and used in a Combined Bayesian Network (CBN). Experiments show that the CBN model can give a health assessment under the proposed definition where drives are predicted to fail no later than their actual failure time 70% or more of the time, while maintaining prediction performance standards at least approximately as good as the individual classifiers.

The connection between traffic congestion and regional demand has become a consensus.A reasonable road network inflow control is essential to prevent the occurrence of congestion.With a consideration of deployment of detectors, an improved macroscopic fundamental diagram-based traffic volume dynamic control method via the feedback control is proposed in this study.Based on OD distribution of trips and connectivity of nodes, a novel method is adopted to identify the key nodes, on which the detectors are laid.And Artificial Neural Network is adopted to predict traffic volume of those sections without detectors.As a case study, the proposed methodology is applied to the regional Road network which is located in downtown area of Nanning, China.Adequate survey and analysis are carried out under current road traffic conditions via simulation study.It is proved that the regional traffic volume dynamic control can ensure steady and orderly regional traffic flow, and enhance the mobility in a saturated traffic conditions.

The connection between traffic congestion and regional demand has become a consensus. A reasonable road network inflow control is essential to prevent the occurrence of congestion. With a consideration of deployment of detectors, an improved macroscopic fundamental diagram-based traffic volume dynamic control method via the feedback control is proposed in this study. Based on the origin–destination distribution of trips and connectivity of nodes, a novel method is adopted to identify the key nodes, on which the detectors are laid. An artificial neural network is adopted to predict traffic volume of those sections without detectors. As a case study, the proposed methodology is applied and estimated via simulation study to the regional road network which is located in the downtown area of Nanning, China. Adequate survey and analysis are carried out under the current road traffic conditions. It is proved that the regional traffic volume dynamic control can ensure steady and orderly regional traffic flow, and enhance the mobility during saturated traffic conditions.

Building human-centered intelligent transport systems (ITSs) requires thorough understanding of the diversified driving styles among drivers. In data-driven driving behavior studies, the temporal distance is deemed as an important variable. However, with respect to the driving style analysis, the appropriate temporal distance range has not been clear yet, and little attention has been drawn to the larger temporal distance that may also have a potential effect on driving style. This paper proposes a new three-layer structure of driving style by using the modified latent Dirichlet allocation (mLDA) model. It is found that the results revealed by the mLDA model based on real driving behavior data are able to align themselves with the results from a driving style questionnaire, and some self-reporting bias is uncovered. More comprehensive driving styles are discovered quantitatively, and the appropriate temporal distance range for driving style analysis is determined. The analyzed results indicate that the time-gap range larger than 10 s are still pivotal and the time-gap range below 20 s is a suitable range for driving style analysis.

Some statistical and machine learning methods have been proposed to build hard drive prediction models based on the SMART attributes, and have achieved good prediction performance. However, these models were not evaluated in the way as they are used in real-world data centers. Moreover, the hard drives deteriorate gradually, but these models can not describe this gradual change precisely. This paper proposes new hard drive failure prediction models based on Classification and Regression Trees, which perform better in prediction performance as well as stability and interpretability compared with the state-of the-art model, the Back propagation artificial neural network model. Experiments demonstrate that the Classification Tree (CT) model predicts over 95% of failures at a false alarm rate (FAR) under 0.1% on a real-world dataset containing 25,792 drives. Aiming at the practical application of prediction models, we test them with different drive families, with fewer number of drives, and with different model updating strategies. The CT model still shows steady and good performance. We propose a health degree model based on Regression Tree (RT) as well, which can give the drive a health assessment rather than a simple classification result. Therefore, the approach can deal with warnings raised by the prediction model in order of their health degrees. We implement a reliability model for RAID-6 systems with proactive fault tolerance and show that our CT model can significantly improve the reliability and/or reduce construction and maintenance cost of large-scale storage systems.