Class Boundaries Research Articles

Quantitative Precipitate Classification and Grain Boundary Property Control in Co/Ni-Base Superalloys

A correlative approach is employed to simultaneously assess structure and chemistry of (carbide and boride) precipitates in a set of novel Co/Ni-base superalloys. Structure is derived from electron backscatter diffraction (EBSD) with pattern template matching, and chemistry obtained with energy dispersive X-ray spectroscopy (EDS). It is found that the principal carbide in these alloys is Mo and W rich with the M6C structure. An M2B boride also exhibiting Mo and W segregation is observed at B levels above approximately 0.085 at. pct. These phases are challenging to distinguish in an SEM with chemical information (EDS or backscatter Z-contrast) alone, without the structural information provided by EBSD. Only correlative chemical and structural fingerprinting is necessary and sufficient to fully define a phase. The identified phases are dissimilar to those predicted using ThermoCalc. We additionally perform an assessment of the grain boundary serratability in these alloys, and observe that significant amplitude is only obtained in the absence of pinning intergranular precipitates.

Mitigating Class-Boundary Label Uncertainty to Reduce Both Model Bias and Variance

The study of model bias and variance with respect to decision boundaries is critically important in supervised learning and artificial intelligence. There is generally a tradeoff between the two, as fine-tuning of the decision boundary of a classification model to accommodate more boundary training samples (i.e., higher model complexity) may improve training accuracy (i.e., lower bias) but hurt generalization against unseen data (i.e., higher variance). By focusing on just classification boundary fine-tuning and model complexity, it is difficult to reduce both bias and variance. To overcome this dilemma, we take a different perspective and investigate a new approach to handle inaccuracy and uncertainty in the training data labels, which are inevitable in many applications where labels are conceptual entities and labeling is performed by human annotators. The process of classification can be undermined by uncertainty in the labels of the training data; extending a boundary to accommodate an inaccurately labeled point will increase both bias and variance. Our novel method can reduce both bias and variance by estimating the pointwise label uncertainty of the training set and accordingly adjusting the training sample weights such that those samples with high uncertainty are weighted down and those with low uncertainty are weighted up. In this way, uncertain samples have a smaller contribution to the objective function of the model’s learning algorithm and exert less pull on the decision boundary. In a real-world physical activity recognition case study, the data present many labeling challenges, and we show that this new approach improves model performance and reduces model variance.

Adversarial Discriminative Active Deep Learning for Domain Adaptation in Hyperspectral Images Classification

ABSTRACT Active deep learning (ADL) presents an appropriate solution for hyperspectral images (HSIs) classification based on domain adaptation (DA) with limited labelled samples in the target domain. But some challenges still exist. First, traditional ADL methods only match the feature distributions between the source and target domains globally without considering the decision boundaries between classes, which makes the ambiguous features near land-cover class boundaries and reduces the classification accuracy. Second, in previous ADL settings, a trained classifier is first used to obtain the predictions for the unlabelled data and then a measure is applied to achieve the uncertainty of such a classifier prediction. This two-step approach does not consider unlabelled data in the classifier training, which ignores dealing with noisy and complex data in the target domain. To overcome these issues, we propose adversarial discriminative active deep learning (ADADL), which presents an adversarial model and incorporates two different land-cover classifiers as a discriminator to consider the class boundaries in aligning feature distributions. Furthermore, ADADL combines the entropy measure along with the cross-entropy loss during training to use the information on the unlabelled data of the target domain. The experimental results with two benchmark HSIs show that the proposed ADADL creates robust transferable features far from the original class boundaries and improves the classification accuracy significantly compared to the state-of-art ADL methods even in complex and noisy data.

Boundary graph convolutional network for temporal action detection

Temporal action proposal generation is a fundamental yet challenging to locate the temporal action in untrimmed videos. Although current proposal generation methods can generate the precise boundary of actions, few focus on considering the relation of proposals. In this paper, we propose a unified framework to generate the temporal boundary proposals with a graph convolution network based on the boundary proposals' feature named Boundary Graph Convolutional Network (BGCN). BGCN draws inspiration from boundary methods and uses edge graph convolution relay on the boundary proposals' feature. First, we use a base layer to fusion the two-stream video features to get two-branches of base features. Then the two-branches of base features enter into the same structure of Proposal Features Graph Convolutional Network (PFGCN): Action PFGCN to extract the action classification score and Boundary PFGCN to extract the ending score and staring score. In proposal features graph convolutional network, we first densely sampled the proposals' feature from the video features. We construct a proposal feature graph, where each proposal feature as a node and their relations between proposals' features as an edge with edge convolution for graph convolution. After that, map the relations into a 2D map score. Experiments on popular benchmarks THUMOS14 demonstrate the superiority of BGCN over (44.8% versus 42.8% at tIoU 0.5) the state-of-the-art proposal generator (e.g., G-TAD, TAL-Net, and BMN) at any of tIoU thresholds from 0.3 to 0.7. On ActivityNet1.3, BGCN also got better results. Moreover, BGCN has high efficiency for action detection with less than 2 MB model size and fast inference time.GCN based on boundary generation for densely produce the action proposals Efficient and novel BGCN model has a great capability to learn the proposal features Has a lower model size for temporal action proposals generation Has fast inference time for temporal action proposals generation.

Enigmatic tenebrionoid beetles from mid-Cretaceous Burmese amber (Coleoptera: Tenebrionoidea)

A new tenebrionoid beetle genus with two species, Kulindrobor enigmaticus gen. et sp. nov. and Kulindrobor magnus gen. et sp. nov., is described from mid-Cretaceous Burmese amber. This highly distinctive genus differs from all extant tenebrionoid families. The beetle's unusual elongate and subcylindrical body with a long prothorax and appendages suggest that they may have occupied subcortical spaces. A phylogenetic analysis under maximum parsimony recovered Kulindrobor gen. nov. as sister to a member of the poorly-defined, polyphyletic Melandryidae with low support. Given that the higher classification and family boundaries in Tenebrionoidea still remain to be fully resolved and that Kulindrobor differs profoundly from all recent melandryids, we provisionally place the new genus as family incertae sedis within Tenebrionoidea.

Domain generalization via optimal transport with metric similarity learning

Generalizing knowledge to unseen domains, where data and labels are unavailable, is crucial for machine learning models. We tackle the domain generalization problem to learn from multiple source domains and generalize to a target domain with unknown statistics. The crucial idea is to extract the underlying invariant features across all the domains. Previous domain generalization approaches mainly focused on learning invariant features and stacking the learned features from each source domain to generalize to a new target domain while ignoring the label information, which will lead to indistinguishable features with an ambiguous classification boundary. One possible solution is to constrain the label-similarity when extracting the invariant features and take advantage of the label similarities for class-specific cohesion and separation of features across domains. Therefore we adopt optimal transport with Wasserstein distance, which could constrain the class label similarity, for adversarial training and also further deploy a metric learning objective to leverage the label information for achieving distinguishable classification boundary. Empirical results show that our proposed method could outperform most of the baselines. Furthermore, ablation studies also demonstrate the effectiveness of each component of our method.

Study on offshore seabed sediment classification based on particle size parameters using XGBoost algorithm

Folk's textual classification scheme which is widely used for sediment study operates with the proportions of gravel, sand, silt and clay fractions conventionally. However, dealing with data from different sources usually needs to face missing values that may make the classification difficult. To solve this problem and discover other methods of analyzing the scheme, with samples of offshore seabed sediment, a two-stage model was established to predict a sample's class using the XGBoost algorithm as well as the grain size parameters as input features. The final model was evaluated with quantitative performance measures of recall, precision and F1 score, and by comparing sediment texture maps using the predicted and the actual data. The results show that the model performs well on extraction of sediment samples without gravel fraction, and prediction of classes that have independent characteristics of grain size parameters or samples not near the boundaries of classes in the ternary diagram. The predicted sediment texture is close to the actual and could be reliable due to errors with little impact on further applications. It is demonstrated that the model could be an auxiliary or alternative approach to offshore sediment texture mapping, as well as supplementary to the analysis of sedimentary environment.

Across regional disparities and beyond family ties: A Ghanaian middle class in the making

ABSTRACT Despite its fuzziness, the term middle class has become increasingly attractive in the past two decades, not only among social scientists and market analysts but also as a term of self-description employed by upwardly mobile individuals. Research on middle classes in the Global South, and especially in Africa, confronts us with particular challenges. First, while conventional class theories take the nation-state as the obvious framework for defining class boundaries, African examples point to the importance of sub-national (as well as transnational) dimensions of class formation. Secondly, marriage ties and kin relations of upwardly mobile individuals in African societies often cut across class boundaries, resulting in ‘multi-class’ families and competing loyalties. The paper addresses these challenges by looking at the emergence of a national middle class in Ghana. More specifically, we discuss how social mobility and class formation play out in two different regions of what is today Ghana: a coastal region that since the seventeenth century has been drawn into global networks and was highly stratified by the end of the nineteenth century; and a marginalized savannah region with an egalitarian society until chieftaincy and education were introduced by the British colonizers in the early twentieth century. These regional disparities and resulting sub-national memberships intersect with the emergence of a national middle class. However, we argue that despite distinct regional trajectories, we are currently witnessing a certain ‘synchronisation’ of social stratification and the formation of a broader middle class that understands itself not only in regional, but national terms.

Hyperspectral Image Classification with Localized Graph Convolutional Filtering

The nascent graph representation learning has shown superiority for resolving graph data. Compared to conventional convolutional neural networks, graph-based deep learning has the advantages of illustrating class boundaries and modeling feature relationships. Faced with hyperspectral image (HSI) classification, the priority problem might be how to convert hyperspectral data into irregular domains from regular grids. In this regard, we present a novel method that performs the localized graph convolutional filtering on HSIs based on spectral graph theory. First, we conducted principal component analysis (PCA) preprocessing to create localized hyperspectral data cubes with unsupervised feature reduction. These feature cubes combined with localized adjacent matrices were fed into the popular graph convolution network in a standard supervised learning paradigm. Finally, we succeeded in analyzing diversified land covers by considering local graph structure with graph convolutional filtering. Experiments on real hyperspectral datasets demonstrated that the presented method offers promising classification performance compared with other popular competitors.

Robust Ordinal Regression: User Credit Grading with Triplet Loss-Based Sampling

With the development of social media sites, user credit grading, which served as an important and fashionable problem, has attracted substantial attention from a slew of developers and operators of mobile applications. In particular, multi-grades of user credit aimed to achieve (1) anomaly detection and risk early warning and (2) personalized information and service recommendation for privileged users. The above two goals still remained as up-to-date challenges. To these ends, in this article, we propose a novel regression-based method. Technically speaking, we define three natural ordered categories including BlockList , GeneralList , and AllowList according to users’ registration and behavior information, which preserve both the global hierarchical relationship of user credit and the local coincident features of users, and hence formulate user credit grading as the ordinal regression problem. Our method is inspired by KDLOR ( kernel discriminant learning for ordinal regression ), which is an effective and efficient model to solve ordinal regression by mapping high-dimension samples to the discriminant region with supervised conditions. However, the performance of KDLOR is fragile to the extreme imbalanced distribution of users. To address this problem, we propose a robust sampling model to balance distribution and avoid overfit or underfit learning, which induces the triplet metric constraint to obtain hard negative samples that well represent the latent ordered class information. A step further, another salient problem lies in ambiguous samples that are noises or located in the classification boundary to impede optimized mapping and embedding. To this problem, we improve sampling by identifying and evading noises in triplets to obtain hard negative samples to enhance robustness and effectiveness for ordinal regression. We organized training and testing datasets for user credit grading by selecting limited items from real-life huge tables of users in the mobile application, which are used in similar problems; moreover, we theoretically and empirically demonstrate the advantages of the proposed model over established datasets.

A new dynamic radius SVDD for fault detection of aircraft engine

When using traditional support vector data description (SVDD) to deal with classification problems, low accuracy is often achieved, especially in the case of noise interference. The traditional SVDD adopts a fixed hypersphere radius to set classification boundary, which means that it tends to force all normal samples and outliers to be separated by a fixed radius, which is unreasonable to a certain extent. In order to reduce the impact of this defect, a dynamic radius support vector data description (DR-SVDD) is firstly proposed in this paper, which introduces the idea of angle in the kernel space and flexibly selects the relevant decision radius for each testing sample. Then, the feasibility and effectiveness of the proposed algorithm are verified on several benchmark data sets. Finally, DR-SVDD based fault detection is carried out for a certain turboshaft engine, and the expected results are obtained, which fully demonstrates its effectiveness and robustness.

Socio-epidemiological study of marital violence among women’s attending health care facility of Central Delhi

Background: Marital or domestic violence (DV) is endemic in communities and countries around the world, cutting across class, race, age, religions and national boundaries. It is defined as a pattern of coercive / controlling behaviors and tactics used by one person over another to gain power and control. The study’s objectives were to understand DV prevalence, factors associated with DV, and health and social outcomes of DV among married women.Methods: This was a healthcare facility based cross-sectional study conducted from Jan-Dec 2011among married women attending antenatal clinic and chest / ICTC clinic. The presence of DV was evaluated using a semi-structured interview designed based on the NHFS-3screening tool. The data collected was analyzed using SPSS version 22.Results: A total of 700 subjects were included. 36% of married women reported the presence of DV. The independent risk factors for DV were belonging to non-Hindu religion (OR=1.95, CI:1.02-3.72), belonging to joint family (OR=1.52, CI: 1.08-2.14), having no daughter (OR=1.7, CI: 1.19-2.44), and ever contraceptive use (OR=1.50, CI: 1.06-2.11) in a multivariate logistic regression model. DV’s main cause was misunderstanding (46%) between the victim’s and the perpetrator. The main perpetrator was husband (81%); main support system for victim’s were parents or parental relatives (63%), while the main type of support received was emotional (92%). Headache (78%) was the most common symptom; 70% subjects were unable to carry out household activities, and 24% victim’s relationship with relatives were affected because of DV.Conclusions: The proportion of DV among the married women is high and is a common problem irrespective of the socio-economic and demographic factors. Hence there is no time for complacency and an urgent need for multisectoral coordination for its quick redressal.

A New Oversampling Method Based on the Classification Contribution Degree

Data imbalance is a thorny issue in machine learning. SMOTE is a famous oversampling method of imbalanced learning. However, it has some disadvantages such as sample overlapping, noise interference, and blindness of neighbor selection. In order to address these problems, we present a new oversampling method, OS-CCD, based on a new concept, the classification contribution degree. The classification contribution degree determines the number of synthetic samples generated by SMOTE for each positive sample. OS-CCD follows the spatial distribution characteristics of original samples on the class boundary, as well as avoids oversampling from noisy points. Experiments on twelve benchmark datasets demonstrate that OS-CCD outperforms six classical oversampling methods in terms of accuracy, F1-score, AUC, and ROC.

Feasibility of Objective Seabed Mapping Techniques in a Coastal Tidal Environment (Wadden Sea, Germany)

The growing interest in monitoring the marine environment has strongly encouraged governmental agencies and research institutes to undertake seabed mapping programs and stimulated scientific interest in innovative mapping methods and tools. In this study, object-based image analysis was used to map a very shallow tidal inlet, characterized by high sediment variability and intense morphodynamic processes. The aim was to test the feasibility of reproducible mapping approaches within extended mapping programs of complex coastal areas. The study is based on full-coverage, high-resolution bathymetry and reflectivity, calibrated by means of sediment samples. Seafloor segmentation and classification were based on a cluster analysis performed on reflectivity, slope, and ruggedness. Statistics of clusters were extracted and analysed to identify the optimal number of clusters and evaluate the suitability of the clustering process to differentiate different seabed types. Clusters and samples data were joined to create a training and validation dataset for characterizing the seabed and carrying out an accuracy assessment. Misclassifications were explored and referred to three main reasons: (i) The not-perfect correspondence between sediment boundaries of classification systems and boundaries derived from the clustering process; (ii) the geomorphological features of the seabed; and (iii) the position accuracy of samples. The study contributes to testing of the feasibility of objective methods and highlights the importance of joining acoustic, lithological, and geomorphological analysis. It highlights issues and the need to critically analyse the mapping results and improve the accuracy of collected data.

Shackled by Patriarchy and Poverty: Women's Experiences with Domestic Violence in North Central Timor and North Sumatra

This article discusses how patriarchal elements of society and culture, in conjunction with poverty, is necessary to comprehend the domestic violence experienced by women. This article departs from a qualitative case study of the experiences of women in Taekas Village, North Central Timor, East Nusa Tenggara, and Pondok Batu Village, Labuhanbatu, North Sumatra, and seeks to obtain a comprehensive and in-depth understanding of how patriarchy and poverty contribute to domestic violence. This article emphasizes that, although domestic violence knows no class, religious, or geographical boundaries, rural women who live in poverty are more vulnerable to domestic violence. This article is hoped to shed light on domestic violence in Indonesia, thereby increasing awareness and providing further impetus for eradicating said practice.

Classification of Very-High-Spatial-Resolution Aerial Images Based on Multiscale Features with Limited Semantic Information

Recently, deep learning has become the most innovative trend for a variety of high-spatial-resolution remote sensing imaging applications. However, large-scale land cover classification via traditional convolutional neural networks (CNNs) with sliding windows is computationally expensive and produces coarse results. Additionally, although such supervised learning approaches have performed well, collecting and annotating datasets for every task are extremely laborious, especially for those fully supervised cases where the pixel-level ground-truth labels are dense. In this work, we propose a new object-oriented deep learning framework that leverages residual networks with different depths to learn adjacent feature representations by embedding a multibranch architecture in the deep learning pipeline. The idea is to exploit limited training data at different neighboring scales to make a tradeoff between weak semantics and strong feature representations for operational land cover mapping tasks. We draw from established geographic object-based image analysis (GEOBIA) as an auxiliary module to reduce the computational burden of spatial reasoning and optimize the classification boundaries. We evaluated the proposed approach on two subdecimeter-resolution datasets involving both urban and rural landscapes. It presented better classification accuracy (88.9%) compared to traditional object-based deep learning methods and achieves an excellent inference time (11.3 s/ha).

An RBF-LVQPNN model and its application to time-varying signal classification

A novel technique is proposed for maintaining the diversity of sample features and modeling imbalanced datasets in multi-channel time-varying signal classification. The RBF-LVQPNN consists of a time-varying signal input layer, an RBF process neuron hidden layer, an LVQ competition layer, a pattern layer, and a classifier. Dynamic clustering was used to divide samples in each pattern class into several pattern subclasses with similar features. Typical signal samples in the pattern subclasses are then used as the kernel centers for RBFPNs, to achieve embedding of the diversity signal class features. The output of the RBFPN layer was used as input to the LVQN. In the competition layer, the ‘winning’ neurons were selected to represent a pattern subclass. In the process of reorganizing pattern subclasses into pattern classes, subclass boundaries were combined into irregular class boundaries to reduce the overlap of decision spaces for different pattern classes. The RBF-LVQNN could improve the memory ability of typical signal features and discriminability of signals, realize structural and data constraints of the model, and improve the modeling property of imbalanced datasets. In this paper, the properties of the RBF-LVQPNN are analyzed and a comprehensive training algorithm is established. Experimental validation was performed with classification diagnoses from seven types of cardiovascular diseases based on 12-lead ECG signals. Results demonstrated that the proposed technique significantly improved both classification accuracy and generalizability comparing with other methods in the experiment.

Cognitive Load Estimation From Speech Commands to Simulated Aircraft

This paper investigates and compares methods for cognitive load (CL) estimation from speech. The majority of previous studies of CL estimation used speech collected in laboratory conditions and conventional speech classification methods. Traditionally laboratory speech contains balanced classes that are labeled by a third party after the speech has been collected. In contrast, the speech used in this research was recorded during an experiment focused on human-machine interaction - where spoken commands were used to control simulated aircraft. The speech was labeled using subjective assessments of CL during an experiment that manipulated workload. Current state-of-the-art Convolutional Neural Network (CNN) classification was used for cognitive load estimation and was compared with conventional Support Vector Machine (SVM) and k-Nearest Neighbor (k-NN) classification. Different speaker-dependence models were compared across 2 and 3 classes. In addition, class boundary selection was optimized to reflect the subjective human workload response sigmoidal curve and compared with linear class boundaries. Results for 3-class CL estimation showed that CNN classifiers trained using speech spectrograms for Partially Speaker Dependent (PSD) models using sigmoidal curve class boundaries provided up to 83.7% accuracy. CNN classifiers outperformed baseline SVM and k-NN classifiers (that used acoustic features) on the same dataset by 13.2% and 10.5% respectively. These outcomes indicate that spectrogram-trained CNN classifiers are a worthy consideration in paralinguistic classification problems.

But Could a Dream: Form and Freedom in Gwendolyn Brooks’s Domestic Sonnets

But Could a Dream: Form and Freedom in Gwendolyn Brooks’s Domestic Sonnets Tess Taylor (bio) In her landmark 1970 interview about artistic practice with George Stavros in Madison, Wisconsin, Gwendolyn Brooks said of art: “You just can’t stay in your comfortable old groves. You have to extend yourself. And it’s easier to stay home and drink beer.” In this ideation, Brooks’s version of comfort is “staying home” and her art is “extending yourself,” presumably moving beyond whatever home is. Art asks us to travel; move beyond comfort; unsettle ourselves; realign our familiar. But what does travel mean here, and now? How do we travel? What is unsettlement? The question has new resonance when we consider that so many of Brooks’s poems are actually about literal homes, or about home-like public places. Her poems circle neighborhoods, exteriors, alleys, inhabitants, businesses, but also actually map the inside of homes themselves. In fact, in that same 1970 interview, Brooks described her work this way: “I believe I have written more ‘kitchenette building’ type poems than I have written about birds singing and feeling overwhelmed by grief.” Indeed, rather than leaving “home,” Brooks’s art of unsettlement actually emerges from exploring and voicing home or homes, in making poetic maps of familiar and lived-in spaces. In fact, it’s arguable that Brooks’s project is to explore her own domestic vernacular, to play the very grooves of neighborhoods and people that represent and embody her sense of community, home, and belonging. It’s also important to notice that Brooks’s whole work exists and is framed with contexts of gender and racial and class boundaries that limit the freedoms [End Page 214] to travel, leave, explore, and move away from such “homes.” Brooks often explores the confinements of the places she writes about, whether the ruts of segregation and racism, or the confining experiences of class and gender, or both. The speakers of Brooks’s poems are often reacting to forms of confinement, “involuntary” places where the domestic is a received rather than chosen space. How do you unsettle yourself while staying home? How do you play your life’s given grooves against themselves? In a pandemic when many of us have been asked to “stay home” for nearly a year now, these questions feel newly resonant. I’m about to spend a few moments in the grooves of Brooks’s domestic sonnets, asking how she uses the sonnet form to dramatize domestic containment—fixed enclosures, whether comfortable or not—and also to hint, in the same quarters, at the way the sonnet itself can offer lyric transit, lyric unsettlement, and lyric liberation. I’m interested in how Brooks’s sonnets harness the tension of lyric both to stage protest and to introduce some possibilities of freedom. In a related groove: I speak from a certain frequency. I offer these thoughts on confinement during the ninth month of a pandemic, as I write at 6:30 am in a garage office which is also, I feel I should report, a much-used laundry room. ________ The sonnet, I think, always carries with it the whiff of its courtly history, even when that history is subverted, upended, repurposed. In fact, the chance to subvert the sonnet’s trailing history is also one of its excitements. To reinvent a form successfully is to charge it both forward and backward in time, electrifying it anew. When Terrance Hayes constructs his American Sonnets for My Past and Future Assassin, and his speakers address the changeable and violent force of American racism, we cannot help but overhear the way that Wyatt or Dryden or Shakespeare crafted the echoing mask of the sonnet’s rhymes to address the elusive, changeable lover. In his later address, Hayes [End Page 215] echoes and subverts this tradition of address: We no longer speak to the unseen lover, but to the unseen assassin, who is outer, but also inner—other, but also self. Hayes’s work reminds us at every turn that the courtly lover may be gone, but we now live in an endless fraught dialogue with racism’s destructive recurring force. The very way...

A Semantic Encoding Out-of-Distribution Classifier for Generalized Zero-Shot Learning

Generalized zero-shot learning (GZSL) poses a challenging problem in that it aims to recognize both seen classes that have appeared in the training stage and unseen classes that have not appeared during training. By utilizing a gating mechanism as the binary classifier, gating methods can decompose GZSL into a conventional ZSL problem and a supervision learning task, thereby leading to outstanding performance by GZSL. However, unseen classes contain many confusing visual samples that distribute too close to the seen class boundaries and are prone to misclassification. To solve this problem, we propose a novel semantic encoding out-of-distribution classifier (SE-OOD) for GZSL. Our method first utilizes semantically consistent mapping to project all the visual samples to their corresponding semantic attributes. Then, both the projected visual samples and original semantic attributes are encoded to their latent representations for distribution alignment. After separating the unseen samples from seen samples in the learned latent space, two domain classifiers are adopted to perform ZSL and supervised classification tasks. Extensive experiments are conducted on four benchmarks, and the results show that our proposed SE-OOD can outperform the state-of-the-arts by a large margin.