Search

Automatic classifier accuracy evaluation (ACAEval) on unlabeled test sets is critical for unseen real-world environments. The use of dataset-level regression on synthesized meta-datasets (comprised of many sample sets) has shown promising results for ACAEval. However, the existing meta-dataset for ACAEval is created using simple image transformations such as rotation and background substitution, which can make it difficult to ensure a reasonable distribution shift between the sample set and the test set. When the distribution shift is large, it becomes challenging to estimate the classifier accuracy on the test set using those sample sets. To ensure more robust ACAEval, this paper attempts to customize a meta-dataset in which each sample set has a reasonable distribution shift to the test set. An intra-class cycle-consistent adversarial learning (ICAL) method is introduced to transfer the style of a labeled training set to the style of the test set, by jointly considering the domain shift issue, the label flipping issue (the semantic information may be changed after style transformation), and the diversity of multiple sample sets in the meta-dataset. Experiments validate that under the same experimental setup, our method outperforms the existing ACAEval methods by a good margin, and achieves state-of-the-art performance on several standard benchmark datasets, including digit classification and natural image classification. • Two authors named Yan Huang in pinyin, one PostDoc (1st author), the other an Associate Prof (3rd author). • Paper on ACAEval for classifier accuracy. Uses meta-dataset technique for unlabeled real-world data. • Customized meta-dataset for ACAEval to address distribution shift between samples and test set. • Our sample set considers label flip issue. Introduces random indicator and FD margin loss. • Experiments demonstrate our ACAEval matches existing methods and surpasses dataset-level regression. [ABSTRACT FROM AUTHOR]

• Face expression analysis suffers from insufficiently annotated data. • We introduce the timid semi-supervised strategy based on diversity for face expression analysis. • The solution produces superior results in both expression recognition and action unit estimation. • Simple solutions based on diversity perform better when biased unlabelled data is available. [Display omitted] In the last years, semi–supervised learning has been proposed as a strategy with high potential for improving machine learning capabilities. Face expression recognition may highly benefit from such a technique, as accurate labeling is both difficult and costly, whereas millions of unlabeled images with human faces are available on the Internet, but without annotations. In this paper we evaluate the benefits of semi–supervised learning in the practical scenarios of face expression analysis. Our conclusion is that better performance is indeed achievable, but by methods that put a distinct emphasis on the diversity of exploring patterns in the unlabeled data domain. The evaluation is carried on multiple tasks such as detecting Action Units on EmotioNet, assessing Action Units intensity on the spontaneous DISFA database and, respectively, recognizing expressions on static images acquired in the wild, from the RAF-DB and FER+ databases. We show that, in these scenarios, a so–called timid semi–supervised learner is more robust and achieves higher performance than standard, confident semi–supervised learners. [ABSTRACT FROM AUTHOR]

• We propose the adaptive aggregation-distillation autoencoder for unsupervised anomaly detection, which considers the diversity of normal patterns and provides a strong guarantee for anomaly detection during training sets containing anomalies. • A density-based landmark is designed to represent diverse normal patterns, which can adaptively update the location and quantity of landmarks during training. • An aggregation-distillation mechanism is built upon the landmark selection in respect to landmark-guided convex polygon reconstruction for minimizing the intra-class variation and differentiating normal from abnormal patterns. • We achieve the state-of-the-art performance on standard benchmarks for unsupervised anomaly detection in ten real-world datasets from different application domains. Anomaly detection (AD) has been receiving great attention as it plays a crucial role in many areas of basic research and industrial applications. However, most existing AD methods not only rely on training on normal data, but also ignore the multi-cluster nature of normal and abnormal patterns. To overcome these limitations, this paper proposes a novel method called Adaptive Aggregation-Distillation AutoEncoder (AADAE) for unsupervised anomaly detection. AADAE is built upon the density-based landmark selection in respect to representing diverse normal patterns. During training, AADAE adaptively updates the location and quantity of landmarks. Then, an aggregation-distillation mechanism is constructed: Firstly, it aggregates the latent representations of normal and anomalous to different landmark-guided regions within the convex polygon with landmarks as vertices, which minimizes the intra-class variation and promotes the separability of normal and abnormal samples. Secondly, the distillation mechanism is applied to obtain reliable detection results when there are anomalies in the training set. The aggregation process motivates AADAE to learn the distribution of multi-cluster normal samples with the help of landmarks, which in turn facilitates the distillation process to differentiate normal from anomalies for training. Extensive empirical studies on ten datasets from different application domains demonstrate the efficiency and generalization ability of the method. [ABSTRACT FROM AUTHOR]

• Time series classifier recommendation tackled for the first time in literature. • Novel proposal of landmarkers for time series classification. • Hierarchical inference of meta-targets analysed for the first time in a consistent way. This work addresses time series classifier recommendation for the first time in the literature by considering several recommendation forms or meta-targets: classifier accuracies, complete ranking, top-M ranking, best set and best classifier. For this, an ad-hoc set of quick estimators of the accuracies of the candidate classifiers (landmarkers) are designed, which are used as predictors for the recommendation system. The performance of our recommender is compared with the performance of a standard method for non-sequential data and a set of baseline methods, which our method outperforms in 7 of the 9 considered scenarios. Since some meta-targets can be inferred from the predictions of other more fine-grained meta-targets, the last part of the work addresses the hierarchical inference of meta-targets. The experimentation suggests that, in many cases, a single model is sufficient to output many types of meta-targets with competitive results. [ABSTRACT FROM AUTHOR]

• We propose a novel multinomial-based method to improve the greedy splitting process of decision trees. • We propose a new random forests variant, dubbed multinomial random forest (MRF), based on which we analyze its consistency and privacy-preservation. • Extensive experiments demonstrate that the performance of MRF is on par with Breiman's RF and is better than all existing consistent RF variants. Despite the impressive performance of random forests (RF), its theoretical properties have not been thoroughly understood. In this paper, we propose a novel RF framework, dubbed multinomial random forest (MRF), to analyze its consistency and privacy-preservation. Instead of deterministic greedy split rule or with simple randomness, the MRF adopts two impurity-based multinomial distributions to randomly select a splitting feature and a splitting value, respectively. Theoretically, we prove the consistency of MRF and analyze its privacy-preservation within the framework of differential privacy. We also demonstrate with multiple datasets that its performance is on par with the standard RF. To the best of our knowledge, MRF is the first consistent RF variant that has comparable performance to the standard RF. The code is available at https://github.com/jiawangbai/Multinomial-Random-Forest. [ABSTRACT FROM AUTHOR]

• Present new generalization risk bounds for multi-class domain adaptation. • Proposed a novel learning method based on the risk bounds. • Performed an empirical study on benchmarking data sets. • Considering class-conditional domain invariance delivers better performances. Neural network-based models have recently shown excellent performance in various kinds of tasks. However, a large amount of labeled data is required to train deep networks, and the cost of gathering labeled training data for every kind of domain is prohibitively expensive. Domain adaptation tries to solve this problem by transferring knowledge from labeled source domain data to unlabeled target domain data. Previous research tried to learn domain-invariant features of source and target domains to address this problem, and this approach has been used as a key concept in various methods. However, domain-invariant features do not mean that a classifier trained on source data can be directly applied to target data because it does not guarantee that data distribution of the same classes will be aligned across two domains. In this paper, we present novel generalization upper bounds for domain adaptation that motivates the need for class-conditional domain invariant learning. Based on this theoretical framework, we then propose a class-conditional domain invariant learning method that can learn a feature space in which features in the same class are expected to be mapped nearby. We empirically experimented that our model showed state-of-the-art performance on standard datasets and showed effectiveness by visualization of latent space. [ABSTRACT FROM AUTHOR]

• The first work on dynamic graph convolutional network for multi-video summarization. • Two strategies are proposed to solve the class imbalance problem of the task. • A diversity regularization is designed to encourage a diverse summarization. • The proposed model can generate diverse summaries and achieve state-of-the-art performances. Multi-video summarization is an effective tool for users to browse multiple videos. In this paper, multi-video summarization is formulated as a graph analysis problem and a dynamic graph convolutional network is proposed to measure the importance and relevance of each video shot in its own video as well as in the whole video collection. Two strategies are proposed to solve the inherent class imbalance problem of video summarization task. Moreover, we propose a diversity regularization to encourage the model to generate a diverse summary. Extensive experiments are conducted, and the comparisons are carried out with the state-of-the-art video summarization methods, the traditional and novel graph models. Our method achieves state-of-the-art performances on two standard video summarization datasets. The results demonstrate the effectiveness of our proposed model in generating a representative summary for multiple videos with good diversity. [ABSTRACT FROM AUTHOR]