Domain adaptation is an important but challenging task. Most of the existing domain methods struggle to extract domain-invariant representation on feature space with entangling information and semantic information. Different from previous efforts entangled space, we aim invariant in latent disentangled (DSR) data. In DSR, assume data generation process controlled by two independent sets variables, i.e., variables variables. Under above assumption, employ a variational auto-encoder...

Recent years have witnessed tremendous interest in deep learning on graph-structured data. Due to the high cost of collecting labeled data, domain adaptation is important supervised graph tasks with limited samples. However, current methods are generally adopted from traditional tasks, and properties data not well utilized. For example, observed social networks different platforms controlled only by crowd or communities but also domain-specific policies background noise. Based these we first...

Marginalized stacked denoising autoencoder (mSDA), has recently emerged with demonstrated effectiveness in domain adaptation. In this paper, we investigate the rationale for why mSDA benefits adaptation tasks from perspective of adaptive regularization. Our investigations focus on two types feature corruption noise: Gaussian noise (mSDA g ) and Bernoulli dropout bd ). Both theoretical empirical results demonstrate that successfully boosts performance but fails to do so. We then propose a new...

Heterogeneous domain adaptation needs supplementary information to link up different domains. However, such may not always be available in real cases. In this paper, a new problem setting called hybrid is investigated. It special case of heterogeneous adaptation, which domains share some common features, but also have their own specific features. We leverage upon features instead achieve effective adaptation. propose general feature transfer framework, can using and simultaneously reduce...

Recent studies have revealed that neural-network-based policies can be easily fooled by adversarial examples. However, while most prior works analyze the effects of perturbing every pixel frame assuming white-box policy access, in this article, we take a more restrictive view toward adversary generation—with goal unveiling limits model's vulnerability. In particular, explore minimalistic attacks defining <italic xmlns:mml="http://www.w3.org/1998/Math/MathML"...

In this paper, the images of tunnel surface are obtained by lining rapid inspection system, and crack forest dataset (TCFD) is established. The disaster characteristics cracks analyzed summarized. Solutions segmentation (TCS) method developed for detection recognition on lining. According to image features optical principal equipment, effective pre-processing steps carried out before extraction. TCFD divided into appropriate number blocks magnify local cracks. Local threshold used traverse...

Imbalanced learning (IL), i.e., unbiased models from class-imbalanced data, is a challenging problem. Typical IL methods including resampling and reweighting were designed based on some heuristic assumptions. They often suffer unstable performance, poor applicability, high computational cost in complex tasks where their assumptions do not hold. In this paper, we introduce novel ensemble framework named MESA. It adaptively resamples the training set iterations to get multiple classifiers...

Reducing domain divergence is a key step in transfer learning. Existing works focus on the minimization of global divergence. However, two domains may consist several shared subdomains, and differ from each other subdomain. In this article, we take local subdomains into account transfer. Specifically, propose to use low-dimensional manifold represent subdomain, align data distribution discrepancy across domains. A maximum mean (M3D) developed measure manifold. We then general framework,...

Unsupervised cross-lingual speech representation learning (XLSR) has recently shown promising results in recognition by leveraging vast amounts of unlabeled data across multiple languages. However, standard XLSR model suffers from language interference problem due to the lack specific modeling ability. In this work, we investigate adaptive training on models. More importantly, propose a novel pretraining approach based sparse sharing sub-networks. It makes room for pruning out unimportant...

Multi-source transfer regression is a practical and challenging problem where capturing the diverse relatedness of different domains key adaptive knowledge transfer. In this article, we propose an effective way explicitly modeling domain each pair through kernel learning. Specifically, first discuss advantages disadvantages existing kernels in handling multi-source problem. To cope with limitations kernels, further novel kms. The proposed kms assigns learnable parametric coefficient to model...

Mass alkaline magmatic activities in Western Shandong during the late Mesozoic controlled mineralization processes of gold and rare earth element (REE) polymetallic deposits region. The Chishan complex is closely associated with REE deposit, which, as third largest light deposit China following Baiyenebo (Inner Mongolia) Mianning (Sichuan) deposits, considered a typical example rock throughout North Craton. To determine how interact each other, systematic study was conducted on petrology,...

In a Brain-Computer Interface (BCI) system, the variations of amplitude and phase in EEG signal convey subjects' movement intention underpin differentiation various mental tasks. Combining these two kinds information under uniform feature extraction framework can better reflect brain states potentially contribute to BCI classification. Here Common Spatial Pattern (CSP) Phase Locking Value (PLV) were used capture information. To integrate procedures, Empirical Mode Decomposition (EMD) is...

Transfer covariance functions, which can model domain similarities and adaptively control the knowledge transfer across domains, are widely used in Gaussian process (GP) based learning. We focus on regression problems a black-box learning scenario, study family of rather general T_*, that similarity heterogeneity domains through multiple kernel A necessary sufficient condition (i) validates GPs using T_* for any data (ii) provides semantic interpretations is given. Moreover, building this...

Intent detection and slot filling are recognized as two very important tasks in a spoken language understanding (SLU) system. In order to model these at the same time, many joint models based on deep neural networks have been proposed recently archived excellent results. addition, graph network has made good achievements field of vision. Therefore, we combine advantages propose new with wheel-graph attention (Wheel-GAT), which is able interrelated connections directly for single intent...

Transfer regression is a practical and challenging problem with important applications in various domains, such as engineering design localization. Capturing the relatedness of different domains key adaptive knowledge transfer. In this paper, we investigate an effective way explicitly modelling domain through transfer kernel, transfer-specified kernel that considers information covariance calculation. Specifically, first give formal definition introduce three basic general forms well cover...

The Shanzhuang banded iron formation (BIF) occurs in the Shancaoyu Formation, Taishan Group, western Shandong Province (WSP), eastern North China Craton (NCC). We constrained BIF depositional age of ∼2.50 Ga based on zircon U-Pb dating leptynite interlayers and cross-cutting metagranitoid. After deposition, was subsequently metamorphosed into amphibolite facies at 2.50–2.45 Ga. According to different mineral assemblages, Fe ores can be divided three types: magnetite + quartz ± hornblende...