Current deep neural networks (DNNs) can easily overfit to biased training data with corrupted labels or class imbalance. Sample re-weighting strategy is commonly used alleviate this issue by designing a weighting function mapping from loss sample weight, and then iterating between weight recalculating classifier updating. approaches, however, need manually pre-specify the as well its additional hyper-parameters. It makes them fairly hard be generally applied in practice due significant...

One of the central tasks genome research is to predict phenotypes and discover some important gene biomarkers. However, there are three main problems in analyzing genomics data marker selection. Such as large p small n, low reproducibility selected biomarkers, high noise. To provide a unified solution alleviate mentioned above, we propose self-paced learning L 1 / 2 ${{\rm{L}}}_{1/2}$ absolute network-based logistic regression model, called SLNL. Through ${L}_{1/2}$ regularization, model can...

Multi-omics data integration is a promising field combining various types of omics data, such as genomics, transcriptomics, and proteomics, to comprehensively understand the molecular mechanisms underlying life disease. However, inherent noise, heterogeneity, high dimensionality multi-omics present challenges for existing methods extract meaningful biological information without overfitting. This paper introduces novel Multi-Omics Meta-learning Algorithm (MUMA) that employs self-adaptive...

Multi-label learning focuses on the ambiguity at label side, i.e., one instance is associated with multiple class labels, where logical labels are always adopted to partition into relevant and irrelevant rigidly. However, relevance or irrelevance of each corresponding essentially relative in real-world tasks distribution more fine-grained than by denoting a certain number description degrees all labels. As not explicitly available most training sets, process named enhancement emerges recover...

Modern deep neural networks can easily overfit to biased training data containing corrupted labels or class imbalance. Sample re-weighting methods are popularly used alleviate this bias issue. Most current methods, however, require manually pre-specify the weighting schemes relying on characteristics of investigated problem and data. This makes them fairly hard be generally applied in practical scenarios, due their significant complexities inter-class variations bias. To address issue, we...

Recent deep neural networks (DNNs) can easily overfit to biased training data with noisy labels. Label correction strategy is commonly used alleviate this issue by identifying suspected labels and then correcting them. Current approaches corrupted usually need manually pre-defined label rules, which makes it hard apply in practice due the large variations of such manual strategies respect different problems. To address issue, we propose a meta-learning model, aiming at attaining an automatic...

Service science research seeks to improve the productivity and quality of service offerings by creating new innovations, facilitating business management, applying practical applications. Recent trends seek apply extend principles from product family design mass customization into development. Product is a cost-effective way achieve allowing highly differentiated products be developed common platform while targeting individual distinct market segments. This article extends concepts...

As a promising area in artificial intelligence, new learning paradigm, called Small Sample Learning (SSL), has been attracting prominent research attention the recent years. In this paper, we aim to present survey comprehensively introduce current techniques proposed on topic. Specifically, SSL can be mainly divided into two categories. The first category of approaches "concept learning", which emphasizes concepts from only few related observations. purpose is simulate human behaviors like...

A game theoretic pricing mechanism for statistically guaranteed service in packet-switched networks is proposed. The provides congestion control, differentiated qualities of service, and efficient resource allocation. For users, the offers better quality lower price. Service providers can base new revenue models within mechanism. We apply this to Internet.

Vascular dementia (VaD) is considered to be the second most common form of after Alzheimer's disease, and no specific drugs have been approved for VaD treatment. We aimed identify shared transcriptomic signatures between frontal cortex temporal in by bioinformatics analyses. Gene ontology pathway enrichment analyses, protein-protein interaction (PPI) hub gene identification, gene-transcription factor interaction, gene-microRNA gene-drug analyses were performed. identified 159 overlapping...

To discover intrinsic inter-class transition probabilities underlying data, learning with noise has become an important approach for robust deep on corrupted labels. Prior methods attempt to achieve such knowledge by pre-assuming strongly confident anchor points 1-probability belonging a specific class, generally infeasible in practice, or directly jointly estimating the matrix and classifier from noisy samples, always leading inaccurate estimation misguided wrong annotation information...

Alzheimer's disease (AD) and type 2 diabetes (T2D) are common in the general elderly population, conferring heavy individual, social, economic stresses on families society. Accumulating evidence indicates T2D to be a risk factor for AD. However, underlying mechanisms this association largely unknown. This study aimed identify shared molecular signatures between AD through integrated analysis of temporal cortex gene expression data. Gene Ontology (GO) pathway enrichment analysis, protein...

The widespread applications in microarray technology have produced the vast quantity of publicly available gene expression datasets. However, analysis data using biostatistics and machine learning approaches is a challenging task due to (1) high noise; (2) small sample size with dimensionality; (3) batch effects (4) low reproducibility significant biomarkers. These issues reveal complexity data, thus significantly obstructing clinical applications. integrative offers an opportunity address...

This paper introduces a 'simulating learning methodology' (SLeM) approach for the methodology determination in general and Auto6 ML particular, reports SLeM framework, approaches, algorithms applications.

Robust loss minimization is an important strategy for handling robust learning issue on noisy labels. Current approaches designing losses involve the introduction of noise-robust factors, i.e., hyperparameters, to control trade-off between noise robustness and learnability. However, finding suitable hyperparameters different datasets with labels a challenging time-consuming task. Moreover, existing methods usually assume that all training samples share common which are independent instances....

Robust loss minimization is an important strategy for handling robust learning issue on noisy labels. Current functions, however, inevitably involve hyperparameter(s) to be tuned, manually or heuristically through cross validation, which makes them fairly hard generally applied in practice. Besides, the non-convexity brought by as well complicated network architecture it easily trapped into unexpected solution with poor generalization capability. To address above issues, we propose a...

A class-aware sample weighting algorithm is developed for general label noise problems. The can effectively tackle complicated and diverse noisy tasks, winning the Championship of 'Arena Contest' Track 1 2022 Greater BayArea (Huangpu) International Algorithm Case Competition.

With the rapid development of Internet, network has been expanding. In order to allow operators provide better quality service, but also effective supervision and management network, identify traffic type classification technology become a hot topic in recent years. this paper, method based on deep learning is proposed. Compared with traditional machine method, accuracy improved obviously.

Person reidentification (Re-ID) aims at matching images of the same identity captured from disjoint camera views, which remains a very challenging problem due to large cross-view appearance variations. In practice, mainstream methods usually learn discriminative feature representation using deep neural network, needs number labeled samples in training process. this article, we design simple yet effective multinetwork collaborative learning (MCFL) framework alleviate data annotation...

Self-paced learning (SPL) is a regime, inspired by human and animal processes, that gradually incorporates simple to more complex samples into training dataset. Recently, SPL has seen significant research progress. However, current algorithms still have critical limitations, such as how determine the age hyper-parameters (especially parameters). Some heuristic strategies based on cross-validation been designed. In addition, setting these parameters manually proposed. are very inefficient,...

This study designs and proposes a method for evaluating the configuration of energy storage integrated renewable generation plants in power spot market, which adopts two-level optimization model "system simulation + plant optimization". The first step is simulation": using market to obtain initial nodal marginal price curtailment plant. second "plant optimization": operation optimize charge-discharge storage. In third step, simulation" conducted again, combined inside brought into system...