Nonnegative matrix factorization (NMF) is a powerful decomposition technique that approximates nonnegative by the product of two low-rank factors. It has been widely applied to signal processing, computer vision, and data mining. Traditional NMF solvers include multiplicative update rule (MUR), projected gradient method (PG), least squares (PNLS), active set (AS). However, they suffer from one or some following three problems: slow convergence rate, numerical instability nonconvergence. In...

Nonnegative matrix factorization (NMF) has become a popular dimension-reduction method and been widely applied to image processing pattern recognition problems. However, conventional NMF learning methods require the entire dataset reside in memory thus cannot be large-scale or streaming datasets. In this paper, we propose an efficient online RSA-NMF algorithm (OR-NMF) that learns incremental fashion solves problem. particular, OR-NMF receives one sample chunk of samples per step updates...

In this paper, we present a non-negative patch alignment framework (NPAF) to unify popular matrix factorization (NMF) related dimension reduction algorithms. It offers new viewpoint better understand the common property of different NMF Although multiplicative update rule (MUR) can solve NPAF and is easy implement, it converges slowly. Thus, propose fast gradient descent (FGD) overcome aforementioned problem. FGD uses Newton method search optimal step size, thus faster than MUR. Experiments...

Non-negative matrix factorization (NMF) approximates a non-negative $X$ by product of two low-rank factor matrices $W$ and $H$. NMF its extensions minimize either the Kullback-Leibler divergence or Euclidean distance between $W^T H$ to model Poisson noise Gaussian noise. In practice, when distribution is heavy tailed, they cannot perform well. This paper presents Manhattan (MahNMF) which minimizes for modeling tailed Laplacian Similar sparse decompositions, MahNMF robustly estimates part...

Non-negative matrix factorization (NMF) minimizes the Euclidean distance between data and its low rank approximation, it fails when applied to corrupted because loss function is sensitive outliers. In this paper, we propose a Truncated CauchyNMF that handle outliers by truncating large errors, develop robustly learn subspace on noisy datasets contaminated We theoretically analyze robustness of comparing with competing models prove has generalization bound which converges at rate order...

Remote sensing image scene classification (RSI-SC) is crucial for various high-level applications, including RSI retrieval, captioning, and object detection. Deep learning-based methods can accurately predict categories. However, these approaches often require numerous labeled samples training, limiting their practicality in real-world RS applications with scarce label resources. In contrast, few-shot remote (FS-RSI-SC) has garnered substantial research interest owing to its potential...

Studies on nowadays human-machine interface have demonstrated that visual information can enhance speech recognition accuracy especially in noisy environments. Deep learning has been widely used to tackle such audio (AVSR) problem due its astonishing achievements both and image recognition. Although existing deep models succeed incorporate into recognition, none of them simultaneously considers sequential characteristics modalities. To overcome this deficiency, we proposed a multimodal...

Drug repositioning, finding new indications for existing drugs, has gained much recent attention as a potentially efficient and economical strategy accelerating therapies into the clinic. Although improvement in sensitivity of computational drug repositioning methods identified numerous credible opportunities, few have been progressed. Arguably "black box" nature action indication is one main blocks to progression, highlighting need that inform on broader target mechanism disease context.We...

Few-shot learning is an important and challenging research topic for remote sensing image scene classification. Many existing approaches address this challenge by using meta-learning metric-learning techniques, which aim to develop feature extractors that can quickly adapt new tasks with limited labeled data. However, these methods are unsuitable real-world datasets have class confusion, high inter-class similarity intra-class diversity. To overcome limitation, we propose a novel effective...

RNA-sequencing is rapidly becoming the method of choice for studying full complexity transcriptomes, however with increasing dimensionality, accurate gene ranking increasingly challenging. This paper proposes an and sensitive that implements discriminant non-negative matrix factorization (DNMF) RNA-seq data. To best our knowledge, this first work to explore utility DNMF ranking. When incorporating Fisher's criteria setting reduced dimension as two, learns two factors approximate original...

Vision Transformer (ViT) models have recently emerged as powerful and versatile tools for various visual tasks. In this article, we investigate ViT in a more challenging scenario within the context of few-shot conditions. Recent work has achieved promising results image classification by utilizing pre-trained vision transformer models. However, employs full fine-tuning downstream tasks, leading to significant overfitting storage issues, especially remote sensing domain. order tackle these...

The recent advance has shown that few-shot learning may be a promising way to alleviate the data reliance of remote sensing image scene classification. However, most existing works focus on extracting distinguishable features only from visual modality, while problem knowledge multiple modalities barely been visited. In this work, we propose text-aware framework for classification (TeAw). Specifically, TeAw converts class names more detailed text descriptions and extracts using pre-trained...

Projective non-negative matrix factorization (PNMF) projects high-dimensional examples X onto a lower-dimensional subspace spanned by basis W and considers WT as their coefficients, i.e., X≈WWT X. Since PNMF learns the natural parts-based representation Wof X, it has been widely used in many fields such pattern recognition computer vision. However, does not perform well classification tasks because completely ignores label information of dataset. This paper proposes Discriminant method...

Advances in DNA microarray technologies have made gene expression profiles a significant candidate identifying different types of cancers. Traditional learning-based cancer identification methods utilize labeled samples to train classifier, but they are inconvenient for practical application because labels quite expensive the clinical research community. This paper proposes semi-supervised projective non-negative matrix factorization method (Semi-PNMF) learn an effective classifier from both...

Non-negative matrix factorization (NMF) approximates a non-negative by the product of two low-rank matrices and achieves good performance in clustering. Recently, semi-supervised NMF (SS-NMF) further improves incorporating part labels few samples into NMF. In this paper, we proposed novel graph based SS-NMF (GSS-NMF). For each sample, GSS-NMF minimizes its distances to same labeled maximizes against different incorporate discriminative information. Since both unlabeled are embedded reduced...

Big models, large datasets, and self-supervised learning (SSL) have recently gained substantial research interest due to their potential alleviate our reliance on annotations. Considering the current high generalization ability of models in literature, we explore letter how helpful SSL can be for a crucial task remote sensing (RS), image scene classification, when forced rely only few labeled samples. We proposed simple prototype-based classification procedure without training fine-tuning,...

Canonical correlation analysis (CCA) has been widely used in pattern recognition and machine learning. However, both CCA its extensions sometimes cannot give satisfactory results. In this paper, we propose a new CCA-type method termed sparse representation based discriminative (SPDCCA) by incorporating information simultaneously into traditional CCA. particular, SPDCCA not only preserves the reconstruction relationship within data on representation, but also maximum-margin information, thus...

Regarding the non-negativity property of magnitude spectrogram speech signals, nonnegative matrix factorization (NMF) has obtained promising performance for separation by independently learning a dictionary on signals each known speaker. However, traditional NM-F fails to represent mixture accurately because dictionaries speakers are learned in absence signals. In this paper, we propose new transductive NMF algorithm (TNMF) jointly learn both speaker and be separated. Since TNMF learns more...

Semi-supervised clustering aims at boosting the performance on unlabeled samples by using labels from a few labeled samples. Constrained NMF (CNMF) is one of most significant semi-supervised methods, and it factorizes whole dataset constrains those same class to have identical encodings. In this paper, we propose novel soft-constrained (SCNMF) method softening hard constraint in CNMF. Particularly, SCNMF into two lower-dimensional factor matrices multiplicative update rule (MUR). To utilize...

Projective non-negative matrix factorization (P-NMF) projects a set of examples onto subspace spanned by basis whose transpose is regarded as the projection matrix. Since PNMF learns natural parts-based representation, it has been successfully used in text mining and pattern recognition. However, non-trivial to analyze convergence optimization algorithms for because its objective function non-convex. In this paper, we propose Box-constrained (BPNMF) method overcome deficiency PNMF....