Search

We propose a simple regularization scheme to handle the problem of mode missing and unstable training in the generative adversarial networks (GAN). The key idea is to utilize the visual features learned by the discriminator. We reconstruct the real data by feeding the generator with the real data features extracted by the discriminator. A reconstruction loss is added to the GAN’s objective function to enforce the generator can reconstruct from the features of the discriminator, which helps to explicitly guide the generator towards to near the probable configurations of real data. The proposed reconstruction loss improves the performance of GAN, produces higher quality images on different dataset, and can be easily combined with other regularization loss functions such as gradient penalty to improve the performance of various GANs. We conducted experiments on the widespread adopted architecture DCGAN and the complicated ResNet architecture across different datasets, the results of which show the effectiveness and robustness of our proposed method.

A novel dynamic refinement approach integrated with constraints and contexts is proposed.Refinement principles & rules based on five defined refinement methods are proposed.ARG & refinement cost are presented to find the potential optimal refinement path.As a further implementation, a general refinement algorithm is described.The results of a case study and the prototype system indicate the effectiveness. Flexible activity refinement plays an important role in improving process flexibility and addressing uncertainties of business processes. However, it is still a challenge to refine flexible activities, and the existing researches on flexible activity refinement such as the refinement principles and methods, and their combination with factors such as constraints and contexts is still lacking. Aiming at this, a novel dynamic refinement approach for flexible activities is proposed, which combines both vertical decomposition and horizontal extension refinements, with the impact of constraints and contexts considered. In particular, we summarize five typical refinement categories, and present a set of activity refinement rules based on them. Furthermore, the decomposition refinement, including the activity decomposition principles, the related rules for trigger event delivering and execution condition transferring is discussed in detail. The extension refinement, which realizes the horizontal refinement, can be integrated with other kinds of refinement and uses constraints to specify activity selection, activity temporal relationships, etc. Then, a tree-like activity refinement graph (ARG) is proposed to represent the refinement process, based on which the refinement cost and refinement degree can be computed to benefit the finding of the potential optimal refinement path. As a further implementation of the proposed refinement approach, a general refinement algorithm is described. Finally, a case study of urolithiasis therapy process and its application are given, and the results indicate the effectiveness of our proposals.

The method of traditional fire detection has many deficiencies. Not only being susceptible to been interference by environmental factors, but also not to find a lot of information timely and effectively from the fire inspection information. At the same time also not to record the specific details when the fire breaks out. These make it difficult for subsequent fire cause investigation. Therefore, this paper puts forward a new method for fire detection and identification by using visual attention mechanism. That is, through simulation of the human visual system, visual attention mechanism can help us from a lot of complex images to quickly find the information “been worth noting” of the image to find critical information, and to eliminate a lot of useless information. The method can significantly speed up image processing and improve the accuracy of image recognition. Several experiments have been designed to verify the effectiveness of the method by using SVM learning and training. The experimental results denote that the novel algorithms based on visual attention mechanism and been mixed linear function nuclear and radial basis function nuclear improve the accuracy and speed of recognition, significantly reduces the false negative rate of fire recognition and improves the accuracy of fire detection.

To solve the problem which unsupervised clustering algorithm is sensitive to parameter settings, the paper proposes a novel heuristic dual-ant clustering algorithm, some problems such as cluster dispersion and over many outliers which exists in traditional algorithm are resolved by adding a new kind of Maintenance Ants. The paper also propose novel heuristic functions to measures the instances similarity and to control the ant movement. Compared with other clustering algorithms, our algorithm do not need to know the number of clustering in advance, the dataset can be automatically clustered in the case of no prior knowledge, it is very suitable for intrusion anomaly detection based on unsupervised clustering. In experiments on network intrusion dataset, our algorithm is compared with the advanced cluster-based anomaly detection algorithm FindCBLOF, without knowing the original partition information of dataset, the experimental results is significantly better than FindCBLOF. It proved our algorithm has a good application value in network intrusion detection.