Search

Adversarial training has become the mainstream method to boost adversarial robustness of deep models. However, it often suffers from the trade-off dilemma, where the use of adversarial examples hurts the standard generalization of models on natural data. To study this phenomenon, we investigate it from the perspective of spatial attention. In brief, standard training typically encourages a model to conduct a comprehensive check to input space. But adversarial training often causes a model to overly concentrate on sparse spatial regions. This reduced tendency is beneficial to avoid adversarial accumulation but easily makes the model ignore abundant discriminative information, thereby resulting in weak generalization. To address this issue, this paper introduces an Attention-Enhanced Learning Framework (AELF) for robustness training. The main idea is to enable the model to inherit the attention pattern of standard pre-trained model through an embedding-level regularization. To be specific, given a teacher model built on natural examples, the embedding distribution of teacher model is used as a static constraint to regulate the embedding outputs of the objective model. This design is mainly supported with that the embedding feature of standard model is usually recognized as a rich semantic integration of input. For implementation, we present a simplified AELFs that can achieve the regularization with single cross entropy loss via the parameter initialization and parameter update strategy. This avoids the extra consistency comparison operation between embedding vectors. Experimental observations verify the rationality of our argument, and experimental results demonstrate that it can achieve remarkable improvements in generalization under the high-level robustness. [ABSTRACT FROM AUTHOR]

Sparse representation-based classification (SRC) has been a breakthrough of face recognition and signal reconstruction recently. However, few face images from the same subject provide insufficient observations. Meanwhile, owing to uncertainty of training images with variations in the appearance of facial illumination, pose and facial expression, it is difficult to successfully address the issues of image recognition and signal reconstruction. Thus, how to construct an optimized comprehensive dictionary is still a non-trivial task. In this paper, we develop a generalized virtual extended dictionary of SRC (V-SRC) to deal with data uncertainty problem, establishing a weighted mechanism which assigns different coefficients between arbitrary two within-class samples to obtain the virtual samples. More specifically, we firstly construct a set of synthesized training samples by means of weighted combination of existing training samples. Secondly, we combine the original and synthesized training samples as an initial extended dictionary. Thirdly, in order to decrease the constructed error in sparse representation, we exploit an elimination scheme to gain the robust training samples with respect to the initial extended training dictionary. The final goal of the proposed method is to utilize the linear combination of competitive training samples to perform face classification. Experimental results conducted on the AR, FERET and Georgia Tech databases demonstrate the effectiveness of the proposed method especially in the case of small sample size problem. [ABSTRACT FROM AUTHOR]

The changes of face images with poses and polarized illuminations increase data uncertainty in face recognition. In fact, synthesized mirror samples can be recognized as representations of the left-right deflection of poses or illuminations of the face. Symmetrical face images generated from the original face images also provide more observations of the face which is useful for improving the accuracy of face recognition. In this paper, to the best of our knowledge, it is the first time that the well-known minimum squared error classification (MSEC) algorithm is used to perform face recognition on an extended face database using synthesized mirror training samples, which is titled as extended minimum squared error classification (EMSEC). By modifying the MSE classification rule, we append the mirror samples to the training set for gaining better classification performance. First, we merge original training samples and mirror samples synthesized from original training samples per subject as mixed training samples. Second, EMSEC algorithm exploits mixed training samples to obtain the projection matrix that can best transform the mixed training samples into predefined class labels. Third, the projection matrix is exploited to simultaneously obtain transform results of the test sample and its nearest neighbor from the mixed training sample set. Finally, we ultimately classify the test sample by combining the transform results of the test sample and the nearest neighbor. As an extension of MSEC, EMSEC reduces the uncertainty of the face observation by auxiliary mirror samples, so that it has better robustness classification performance than traditional MSEC. Experimental results on the ORL, GT, and FERET databases show that EMSEC has better generalization ability than traditional MSEC. [ABSTRACT FROM AUTHOR]

Abstract: According to the characteristics of offshore wells using Electrical Submersible Pump (ESP), the coupled calculating method is adopted to establish the flow and heat transfer model of wellbore. The flow and heat transfer model is compiled by C language and validated by field measurements. The numerical calculation results show that the factors of marine environment, heat emitted by pump and cable, and the Joule–Thomson effect have significant impact on the wellbore temperature and pressure distribution. When temperature decreases on the condition of low water cut and high gas–oil ratio, the increase of gas–liquid ratio and decrease of density will retard the pressure reduction. However, when the temperature decreases on the condition of high water cut and low gas–oil ratio, decrease of the gas–liquid ratio and increase of density will accelerate the pressure reduction. [Copyright &y& Elsevier]

Probability density machine (PDM) is a novel algorithm which was proposed recently for addressing class imbalance learning (CIL) problem. PDM can capture priori data distribution information well and present robust performance in various CIL applications. However, we also note that the PDM is sensitive to CIL data with varying density and/or small disjunctions which means there are two or multiple obvious sub-clusters within the same class, as on this kind of data, the estimation of conditional probability might be extremely inaccurate. To address this problem, we introduce the shared nearest neighbors (SNN) clustering technique into PDM procedure and propose a novel SNN-PDM algorithm. Specifically, the SNN can adapt varying density and capture small disjunctions existing in data distribution well. We evaluate the proposed algorithm on a large amount of CIL datasets, and the results show that the proposed SNN-PDM algorithm outperforms the PDM and several previous methods. Meanwhile, in comparison with PDM, the SNN-PDM has less time consumption. [ABSTRACT FROM AUTHOR]

An erosion often occurs, when brazing cemented carbide with Cu-Mn based filler alloy, therefore, the binder phase erosion and interface evolution are systematically investigated. When heating, the binder phase Co dissolves into the molten filler and the erosion product acquires liquid state. Heavy WC particles successively submerge into the molten filler, while the erosion layer becomes incompact. With increasing brazing temperature, the tensile strength increases to the maximum 589MPa at 970 °C with holding for 300 sec, and then decreases to 431 MPa at 1010 °C. The microscopic fracture morphology analysis reveals that the erosion product possesses poor wettability against WC particles. When long-time heat preservation at 970 °C, a structure gradient layer is locally formed, because the de-bonded WC particles submerge into the liquid filler metal successively, which promotes the thermal stress release and obtaining high-strength joint. [ABSTRACT FROM AUTHOR]