This paper presents a novel probabilistic approach to fusing multimodal metadata for event based home photo clustering. Photo events are characterized by the coherence of multimodality including time, content and camera settings. We incorporate these multimodal metadata into a unified probabilistic framework, in which event is taken as a latent semantic concept and discovered by fitting a generative model through an Expectation-Maximization (EM) algorithm. This approach is general and unsupervised, without any training procedure or predefined threshold. The experimental evaluations on 14k photos taken by 10 amateur photographers have indicated the effectiveness and efficiency of the proposed framework in browsing and searching personal photo collections.

No abstract

Weather forecasting is usually solved through numerical weather prediction (NWP), which can sometimes lead to unsatisfactory performance due to inappropriate setting of the initial states. In this paper, we design a data-driven method augmented by an effective information fusion mechanism to learn from historical data that incorporates prior knowledge from NWP. We cast the weather forecasting problem as an end-to-end deep learning problem and solve it by proposing a novel negative log-likelihood error (NLE) loss function. A notable advantage of our proposed method is that it simultaneously implements single-value forecasting and uncertainty quantification, which we refer to as deep uncertainty quantification (DUQ). Efficient deep ensemble strategies are also explored to further improve performance. This new approach was evaluated on a public dataset collected from weather stations in Beijing, China. Experimental results demonstrate that the proposed NLE loss significantly improves generalization compared to mean squared error (MSE) loss and mean absolute error (MAE) loss. Compared with NWP, this approach significantly improves accuracy by 47.76%, which is a state-of-the-art result on this benchmark dataset. The * Jie Lu and Tianrui Li are the correspondence authors of this paper. † Yu Zheng is also affiliated with Xidian University. preliminary version of the proposed method won 2nd place in an online competition for daily weather forecasting 1. CCS CONCEPTS • Applied computing → Environmental sciences.

No abstract

Through analyzing the factors that influence end-point manganese content during BOF steelmaking process, multiple linear regression model for prediction of end-point manganese content was obtained on the basis of actual production data. Given the advantages of artificial neural network, it was used to predict end-point manganese content during BOF steelmaking process, and BP neural network model was established. By means of combining the characteristics of genetic algorithm and BP neural network completely, a combined GA-BP neural network model was established. The verification and comparison of the above three models show that the combined GA-BP neural network model has the highest prediction accuracy. The hit rate of the combined GA-BP neural network model is 90% and 84% respectively when predictive errors of the model are within ±0.03% and ±0.025%. Compared with two models aboved, the combined GA-BP neural network model could provide the most accurate prediction of end-point manganese content, and thus represents a good reference for real production.

No abstract

In the field of transformer failure diagnosis, the potential correlation between different characteristic parameters and failures is difficult to detect using traditional methods. Further, the quantities of inspection data have not been fully utilized. To improve the accuracy of transformer diagnosis, this study establishes a diagnosis model based on fuzzy association rules combined with case-based reasoning (CBR) to evaluate the failure types, fault locations, and cause of breakdown in power transformers. First, the inspection data of transformers are collected from several substations over 10 years. Then, the preprocessed data are randomly separated into training and testing sets. For the training set, fuzzy association rules are built for multiple parameters to narrow the search scope of base case preliminarily. Next, CBR is applied to determine the most similar cases. The failure information of the target transformer can be obtained in detail along with the most similar base case. Finally, the accuracy of the model is validated shown in case studies using the testing data set. The result demonstrates that the diagnosis model provides a higher accuracy than the classic IEC 60599 three-ratio method used in the current industry, which means that this diagnosis model has better performance on diagnosis accuracy.

No abstract