Existing counting models suffer significant performance degradation when tested on data from unknown scenarios (out-of-distribution data), due to the domain shift problem. In practical applications, it is crucial that the model exhibits robust generalization capability without relying on the target domain. This study proposes a novel domain generalization network, GFCount, which involves an innovative meta-learning strategy that leverages graphs to capture the inter-image relevance, thereby facilitating the partitioning of data into pseudo-source and pseudo-target domains. Furthermore, to strengthen the distinction between domain-relevant and domain-invariant features, we devise a feature disentanglement method that integrates style normalization with memory spaces. Finally, we formulate a tailored counting loss function specifically designed for domain generalization, aiming to disentangle features and enhance the quality of the generated density maps. Compared with the advanced approaches, GFCount achieves state-of-the-art results on three public benchmarks, with counting errors on SHB→SHA significantly decreasing by 14.2%.

3D Gaussian Splatting is widely used in 3D reconstruction and has applications in novel view synthesis and scene generation. Recent work has addressed this problem by leveraging multi-view data for high-quality 3D scene reconstruction. Unfortunately, these approaches suffer from overfitting with sparse-view data due to insufficient view information, resulting in artifacts and aliasing. In contrast, we propose a novel method, MPM-GS, which incorporates monocular depth estimation, virtual views, and regularization techniques to address these challenges and improve reconstruction quality under sparse-view conditions. This fixes the overfitting and artifact issues, however, it does not solve all the limitations of sparse-view data in extreme edge cases. Consequently, we develop a novel densification strategy to optimize Gaussian point distributions and improve scene accuracy. While promising, this densification process is non-trivial, as it requires balancing efficiency with rendering fidelity. Therefore, we further optimize Gaussian points and generate new, representative points to enhance both accuracy and computational efficiency. We evaluate MPM-GS both qualitatively and quantitatively on the Tanks & Temples and LLFF datasets, achieving excellent rendering quality and fast rendering speed, particularly in novel view synthesis.

Multi-view dimensionality reduction (MvDR) is a potent approach for addressing the high-dimensional challenges in multi-view data. Recently, contrastive learning (CL) has gained considerable attention due to its superior performance. However, most CL-based methods focus on promoting consistency between any two cross views from the perspective of subspace samples, which extract features containing redundant information and fail to capture view-specific discriminative information. In this study, we propose feature- and recovery-level contrastive losses to eliminate redundant information and capture view-specific discriminative information, respectively. Based on this, we construct a novel MvDR method based on triple contrastive heads (TCH). This method combines sample-, feature-, and recovery-level contrastive losses to extract sufficient yet minimal subspace discriminative information in accordance with the information bottleneck principle. Furthermore, the relationship between TCH and mutual information is revealed, which provides the theoretical support for the outstanding performance of our method. Our experiments on five real-world datasets show that the proposed method outperforms existing methods.

Heterogeneous domain adaptation (HDA) aims to address the transfer learning problems where the source domain and target domain are represented by heterogeneous features. The existing HDA methods based on matrix factorization have been proven to learn transferable features effectively. However, these methods only preserve the original neighbor structure of samples in each domain and do not use the label information to explore the similarity and separability between samples. This would not eliminate the cross-domain bias of samples and may mix cross-domain samples of different classes in the common subspace, misleading the discriminative feature learning of target samples. To tackle the aforementioned problems, we propose a novel matrix factorization-based HDA method called HDA with generalized similarity and dissimilarity regularization (HGSDR). Specifically, we propose a similarity regularizer by establishing the cross-domain Laplacian graph with label information to explore the similarity between cross-domain samples from the identical class. And we propose a dissimilarity regularizer based on the inner product strategy to expand the separability of cross-domain labeled samples from different classes. For unlabeled target samples, we keep their neighbor relationship to preserve the similarity and separability between them in the original space. Hence, the generalized similarity and dissimilarity regularization is built by integrating the above regularizers to facilitate cross-domain samples to form discriminative class distributions. HGSDR can more efficiently match the distributions of the two domains both from the global and sample viewpoints, thereby learning discriminative features for target samples. Extensive experiments on the benchmark datasets demonstrate the superiority of the proposed method against several state-of-the-art methods.

In this paper, we revisit techniques for uncertainty estimation within deep neural networks and consolidate a suite of techniques to enhance their reliability. Our investigation reveals that an integrated application of diverse techniques-spanning model regularization, classifier and optimization-substantially improves the accuracy of uncertainty predictions in image classification tasks. The synergistic effect of these techniques culminates in our novel SURE approach. We rigorously evaluate SURE against the benchmark of failure prediction, a critical testbed for uncertainty estimation efficacy. Our results showcase a consistently better performance than models that individually deploy each technique, across various datasets and model architectures. When applied to real-world challenges, such as data corruption, label noise, and long-tailed class distribution, SURE exhibits remarkable robustness, delivering results that are superior or on par with current state-of-the-art specialized methods. Particularly on Animal-10N and Food-101N for learning with noisy labels, SURE achieves state-of-the-art performance without any task-specific adjustments. This work not only sets a new benchmark for robust uncertainty estimation but also paves the way for its application in diverse, real-world scenarios where reliability is paramount. Our code is available at https://yutingli0606.github.io/SURE/.

While the great capability of Transformers significantly boosts prediction accuracy, it could also yield overconfident predictions and require calibrated uncertainty estimation, which can be commonly tackled by Gaussian processes (GPs). Existing works apply GPs with symmetric kernels under variational inference to the attention kernel; however, omitting the fact that attention kernels are in essence asymmetric. Moreover, the complexity of deriving the GP posteriors remains high for large-scale data. In this work, we propose Kernel-Eigen Pair Sparse Variational Gaussian Processes (KEP-SVGP) for building uncertainty-aware self-attention where the asymmetry of attention kernels is tackled by Kernel SVD (KSVD) and a reduced complexity is acquired. Through KEP-SVGP, i) the SVGP pair induced by the two sets of singular vectors from KSVD w.r.t. the attention kernel fully characterizes the asymmetry; ii) using only a small set of adjoint eigenfunctions from KSVD, the derivation of SVGP posteriors can be based on the inversion of a diagonal matrix containing singular values, contributing to a reduction in time complexity; iii) an evidence lower bound is derived so that variational parameters and network weights can be optimized with it. Experiments verify our excellent performances and efficiency on in-distribution, distribution-shift and out-of-distribution benchmarks.

In contrast with Mercer kernel-based approaches as used e.g. in Kernel Principal Component Analysis (KPCA), it was previously shown that Singular Value Decomposition (SVD) inherently relates to asymmetric kernels and Asymmetric Kernel Singular Value Decomposition (KSVD) has been proposed. However, the existing formulation to KSVD cannot work with infinite-dimensional feature mappings, the variational objective can be unbounded, and needs further numerical evaluation and exploration towards machine learning. In this work, i) we introduce a new asymmetric learning paradigm based on coupled covariance eigenproblem (CCE) through covariance operators, allowing infinite-dimensional feature maps. The solution to CCE is ultimately obtained from the SVD of the induced asymmetric kernel matrix, providing links to KSVD. ii) Starting from the integral equations corresponding to a pair of coupled adjoint eigenfunctions, we formalize the asymmetric Nyström method through a finite sample approximation to speed up training. iii) We provide the first empirical evaluations verifying the practical utility and benefits of KSVD and compare with methods resorting to symmetrization or linear SVD across multiple tasks.

Driven by Industry 4.0, digital twin, as a key enabling technology for digital transformation and intelligent upgrade, has attracted growing attention in various fields of agriculture, such as aquaponics factory. Although the digital twin has made formidable progress in theory and application, there are still many doubts and challenges for aquaponics factory. Based on a review of publications related to digital twin, in this contribution, we summarized the connotation of digital twin, including an overview of the research evolution, architecture, and clarification of digital twin and other concepts, such as cyber-physical systems and simulations. Enabling technologies of digital twin are also introduced from a five-dimensional model perspective. On the basis of maintaining a consensus understanding of digital twin, we explored the potential of digital twin in aquaponics without violating the initial vision of digital twin. First, we provided a comprehensive and insightful summary of aquaponics factory and the application of digital twin in agriculture, with the aim of discovering the possibilities and directions for introducing digital twin into aquaponics factory. Then, we dissected in detail practical cases of digital twin related to the production of aquaponics factory, highlighting the added value that digital twin may bring to aquaponics factory, and explored urgent challenges and research directions.

Supervised learning can be viewed as distilling relevant information from input data into feature representations. This process becomes difficult when supervision is noisy as the distilled information might not be relevant. In fact, recent research shows that networks can easily overfit all labels including those that are corrupted, and hence can hardly generalize to clean datasets. In this article, we focus on the problem of learning with noisy labels and introduce compression inductive bias to network architectures to alleviate this overfitting problem. More precisely, we revisit one classical regularization named Dropout and its variant Nested Dropout. Dropout can serve as a compression constraint for its feature dropping mechanism, while Nested Dropout further learns ordered feature representations with respect to feature importance. Moreover, the trained models with compression regularization are further combined with co-teaching for performance boost. Theoretically, we conduct bias variance decomposition of the objective function under compression regularization. We analyze it for both single model and co-teaching. This decomposition provides three insights: 1) it shows that overfitting is indeed an issue in learning with noisy labels; 2) through an information bottleneck formulation, it explains why the proposed feature compression helps in combating label noise; and 3) it gives explanations on the performance boost brought by incorporating compression regularization into co-teaching. Experiments show that our simple approach can have comparable or even better performance than the state-of-the-art methods on benchmarks with real-world label noise including Clothing1M and ANIMAL-10N. Our implementation is available at https://yingyichen-cyy.github.io/CompressFeatNoisyLabels/.

Recently, a new line of works has emerged to understand and improve self-attention in Transformers by treating it as a kernel machine. However, existing works apply the methods for symmetric kernels to the asymmetric self-attention, resulting in a nontrivial gap between the analytical understanding and numerical implementation. In this paper, we provide a new perspective to represent and optimize self-attention through asymmetric Kernel Singular Value Decomposition (KSVD), which is also motivated by the low-rank property of self-attention normally observed in deep layers. Through asymmetric KSVD, i) a primal-dual representation of self-attention is formulated, where the optimization objective is cast to maximize the projection variances in the attention outputs; ii) a novel attention mechanism, i.e., Primal-Attention, is proposed via the primal representation of KSVD, avoiding explicit computation of the kernel matrix in the dual; iii) with KKT conditions, we prove that the stationary solution to the KSVD optimization in Primal-Attention yields a zero-value objective. In this manner, KSVD optimization can be implemented by simply minimizing a regularization loss, so that low-rank property is promoted without extra decomposition. Numerical experiments show state-of-the-art performance of our Primal-Attention with improved efficiency. Moreover, we demonstrate that the deployed KSVD optimization regularizes Primal-Attention with a sharper singular value decay than that of the canonical self-attention, further verifying the great potential of our method. To the best of our knowledge, this is the first work that provides a primal-dual representation for the asymmetric kernel in self-attention and successfully applies it to modelling and optimization.

The success of Vision Transformer (ViT) in various computer vision tasks has promoted the ever-increasing prevalence of this convolution-free network. The fact that ViT works on image patches makes it potentially relevant to the problem of jigsaw puzzle solving, which is a classical self-supervised task aiming at reordering shuffled sequential image patches back to their original form. Solving jigsaw puzzle has been demonstrated to be helpful for diverse tasks using Convolutional Neural Networks (CNNs), such as feature representation learning, domain generalization and fine-grained classification. In this paper, we explore solving jigsaw puzzle as a self-supervised auxiliary loss in ViT for image classification, named Jigsaw-ViT. We show two modifications that can make Jigsaw-ViT superior to standard ViT: discarding positional embeddings and masking patches randomly. Yet simple, we find that the proposed Jigsaw-ViT is able to improve on both generalization and robustness over the standard ViT, which is usually rather a trade-off. Numerical experiments verify that adding the jigsaw puzzle branch provides better generalization to ViT on large-scale image classification on ImageNet. Moreover, such auxiliary loss also improves robustness against noisy labels on Animal-10N, Food-101N, and Clothing1M, as well as adversarial examples. Our implementation is available at https://yingyichen-cyy.github.io/Jigsaw-ViT.

Feature-based domain adaptation methods project samples from different domains into the same feature space and try to align the distribution of two domains to learn an effective transferable model. The vital problem is how to find a proper way to reduce the domain shift and improve the discriminability of features. To address the above issues, we propose a unified Probability-based Graph embedding Cross-domain and class Discriminative feature learning framework for unsupervised domain adaptation (PGCD). Specifically, we propose novel graph embedding structures to be the class discriminative transfer feature learning item and cross-domain alignment item, which can make the same-category samples compact in each domain, and fully align the local and global geometric structure across domains. Besides, two theoretical analyses are given to prove the interpretability of the proposed graph structures, which can further describe the relationships between samples to samples in single-domain and cross-domain transfer feature learning scenarios. Moreover, we adopt novel weight strategies via probability information to generate robust centroids in each proposed item to enhance the accuracy of transfer feature learning and reduce the error accumulation. Compared with the advanced approaches by comprehensive experiments, the promising performance on the benchmark datasets verify the effectiveness of the proposed model.

In this paper, we attempt to solve a long-lasting open question for non-positive definite (non-PD) kernels in machine learning community: can a given non-PD kernel be decomposed into the difference of two PD kernels (termed as positive decomposition)? We cast this question as a distribution view by introducing the signed measure, which transforms positive decomposition to measure decomposition: a series of non-PD kernels can be associated with the linear combination of specific finite Borel measures. In this manner, our distribution-based framework provides a sufficient and necessary condition to answer this open question. Specifically, this solution is also computationally implementable in practice to scale non-PD kernels in large sample cases, which allows us to devise the first random features algorithm to obtain an unbiased estimator. Experimental results on several benchmark datasets verify the effectiveness of our algorithm over the existing methods.

Analysis and forecasting of sequential data, key problems in various domains of engineering and science, have attracted the attention of many researchers from different communities. When predicting the future probability of events using time series, recurrent neural networks (RNNs) are an effective tool that have the learning ability of feedforward neural networks and expand their expression ability using dynamic equations. Moreover, RNNs are able to model several computational structures. Researchers have developed various RNNs with different architectures and topologies. To summarize the work of RNNs in forecasting and provide guidelines for modeling and novel applications in future studies, this review focuses on applications of RNNs for time series forecasting in environmental factor forecasting. We present the structure, processing flow, and advantages of RNNs and analyze the applications of various RNNs in time series forecasting. In addition, we discuss limitations and challenges of applications based on RNNs and future research directions. Finally, we summarize applications of RNNs in forecasting.

SUMMARY: ODORactor is an open access web server aimed at providing a platform for identifying odorant receptors (ORs) for small molecules and for browsing existing OR-ligand pairs. It enables the prediction of ORs from the molecular structures of arbitrary chemicals by integrating two individual functionalities: odorant verification and OR recognition. The prediction of the ORs for several odorants was experimentally validated in the study. In addition, ODORactor features a comprehensive repertoire of olfactory information that has been manually curated from literature. Therefore, ODORactor may provide an effective way to decipher olfactory coding and could be a useful server tool for both basic olfaction research in academia and for odorant discovery in industry. AVAILABILITY: Freely available at http://mdl.shsmu.edu.cn/ODORactor CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.