Learning From Noisy Labels With Deep Neural Networks: A Survey.

Deep learning has achieved remarkable success in numerous domains with help from large amounts of big data. However, the quality of data labels is a concern because of the lack of high-quality labels in many real-world scenarios. As noisy labels severely degrade the generalization performance of deep neural networks, learning from noisy labels (robust training) is becoming an important task in modern deep learning applications. In this survey, we first describe the problem of learning with label noise from a supervised learning perspective. Next, we provide a comprehensive review of 62 state-of-the-art robust training methods, all of which are categorized into five groups according to their methodological difference, followed by a systematic comparison of six properties used to evaluate their superiority. Subsequently, we perform an in-depth analysis of noise rate estimation and summarize the typically used evaluation methodology, including public noisy datasets and evaluation metrics. Finally, we present several promising research directions that can serve as a guideline for future studies.

High enrichment and near real-time quantification of airborne viruses using a wet-paper-based electrochemical immunosensor under an electrostatic field.

Conventional airborne virus measurement usually requires appreciable sampling and detection times. Viral aerosols should also be collected or prepared in a liquid medium whose volume typically ranges from milliliters to tens of milliliters; hence, many sampling and detection steps need to be taken with the unit horizontal or immobile. Moreover, viral aerosols need to be sufficiently enriched, which makes real-time monitoring difficult. Herein, we present a near real-time enrichment and quantification system of airborne viruses that consists of a wet-paper-based electrochemical immunosensor with a gel electrolyte and a modified electrostatic particle concentrator. A small amount of phosphate-buffered saline flowed on the electrode, which resulted in sensor electrodes that are barely wet (covered in a thin buffer film measuring several micrometers) to ensure antigen-antibody interaction and the removal of non-target particles on the electrode surface. This system ensures that airborne viruses are highly enriched on the working electrode of the immunosensor, and it is possible to measure the MS2 virus particle concentrations every 10 min for 60 min stably and selectively against non-target airborne viruses and bacteria at horizontal and tilted measurement configurations. This system thus has the potential to be used in the real-time mobile monitoring of airborne microorganisms.

Can Robots Help Working Parents with Childcare? Optimizing Childcare Functions for Different Parenting Characteristics.

Is it true that parents always prioritize educational effectiveness when selecting childcare services? The current study identified the potential requirements of dual-income parents toward social robots' diverse childcare functions (e.g., socialization, education, entertainment, and consultation). The results revealed that parental attitudes toward robots were made more positive by all the childcare functions of robots except for their educational features. Furthermore, parents' expectations of childcare functions varied based on their parenting characteristics. Spectral clustering analysis identified distinctive parenting styles (e.g., family-oriented, work-oriented, noninterventional, and dominant), and multigroup structural equation modeling suggested that the impact of robots' socialization function was significant in all parent groups, while other childcare functions exerted limited influence according to specific parenting styles. In addition, children's characteristics were found to alter parents' preferences for each childcare function. These results offer practical implications for the early adoption of childcare robots through predetermining parents' acceptability based on their specific parenting characteristics.

Recent advancements in the measurement of pathogenic airborne viruses.

Air-transmissible pathogenic viruses, such as influenza viruses and coronaviruses, are some of the most fatal strains and spread rapidly by air, necessitating quick and stable measurements from sample air volumes to prevent further spread of diseases and to take appropriate steps rapidly. Measurements of airborne viruses generally require their collection into liquids or onto solid surfaces, with subsequent hydrosolization and then analysis using the growth method, nucleic-acid-based techniques, or immunoassays. Measurements can also be performed in real time without sampling, where species-specific determination is generally disabled. In this review, we introduce some recent advancements in the measurement of pathogenic airborne viruses. Air sampling and measurement technologies for viral aerosols are reviewed, with special focus on the effects of air sampling on damage to the sampled viruses and their measurements. Measurement of pathogenic airborne viruses is an interdisciplinary research area that requires understanding of both aerosol technology and biotechnology to effectively address the issues. Hence, this review is expected to provide some useful guidelines regarding appropriate air sampling and virus detection methods for particular applications.

DTranNER: biomedical named entity recognition with deep learning-based label-label transition model.

BACKGROUND: Biomedical named-entity recognition (BioNER) is widely modeled with conditional random fields (CRF) by regarding it as a sequence labeling problem. The CRF-based methods yield structured outputs of labels by imposing connectivity between the labels. Recent studies for BioNER have reported state-of-the-art performance by combining deep learning-based models (e.g., bidirectional Long Short-Term Memory) and CRF. The deep learning-based models in the CRF-based methods are dedicated to estimating individual labels, whereas the relationships between connected labels are described as static numbers; thereby, it is not allowed to timely reflect the context in generating the most plausible label-label transitions for a given input sentence. Regardless, correctly segmenting entity mentions in biomedical texts is challenging because the biomedical terms are often descriptive and long compared with general terms. Therefore, limiting the label-label transitions as static numbers is a bottleneck in the performance improvement of BioNER. RESULTS: We introduce DTranNER, a novel CRF-based framework incorporating a deep learning-based label-label transition model into BioNER. DTranNER uses two separate deep learning-based networks: Unary-Network and Pairwise-Network. The former is to model the input for determining individual labels, and the latter is to explore the context of the input for describing the label-label transitions. We performed experiments on five benchmark BioNER corpora. Compared with current state-of-the-art methods, DTranNER achieves the best F1-score of 84.56% beyond 84.40% on the BioCreative II gene mention (BC2GM) corpus, the best F1-score of 91.99% beyond 91.41% on the BioCreative IV chemical and drug (BC4CHEMD) corpus, the best F1-score of 94.16% beyond 93.44% on the chemical NER, the best F1-score of 87.22% beyond 86.56% on the disease NER of the BioCreative V chemical disease relation (BC5CDR) corpus, and a near-best F1-score of 88.62% on the NCBI-Disease corpus. CONCLUSIONS: Our results indicate that the incorporation of the deep learning-based label-label transition model provides distinctive contextual clues to enhance BioNER over the static transition model. We demonstrate that the proposed framework enables the dynamic transition model to adaptively explore the contextual relations between adjacent labels in a fine-grained way. We expect that our study can be a stepping stone for further prosperity of biomedical literature mining.

Characterization of a highly pathogenic H5N1 avian influenza A virus isolated from duck meat.

Since the 1997 H5N1 influenza virus outbreak in humans and poultry in Hong Kong, the emergence of closely related viruses in poultry has raised concerns that additional zoonotic transmissions of influenza viruses from poultry to humans may occur. In May 2001, an avian H5N1 influenza A virus was isolated from duck meat that had been imported to South Korea from China. Phylogenetic analysis of the hemagglutinin (HA) gene of A/Duck/Anyang/AVL-1/01 showed that the virus clustered with the H5 Goose/Guandong/1/96 lineage and 1997 Hong Kong human isolates and possessed an HA cleavage site sequence identical to these isolates. Following intravenous or intranasal inoculation, this virus was highly pathogenic and replicated to high titers in chickens. The pathogenesis of DK/Anyang/AVL-1/01 virus in Pekin ducks was further characterized and compared with a recent H5N1 isolate, A/Chicken/Hong Kong/317.5/01, and an H5N1 1997 chicken isolate, A/Chicken/Hong Kong/220/97. Although no clinical signs of disease were observed in H5N1 virus-inoculated ducks, infectious virus could be detected in lung tissue, cloacal, and oropharyngeal swabs. The DK/Anyang/AVL-1/01 virus was unique among the H5N1 isolates in that infectious virus and viral antigen could also be detected in muscle and brain tissue of ducks. The pathogenesis of DK/Anyang/AVL-1/01 virus was characterized in BALB/c mice and compared with the other H5N1 isolates. All viruses replicated in mice, but in contrast to the highly lethal CK/HK/220/97 virus, DK/Anyang/AVL-1/01 and CK/HK/317.5/01 viruses remained localized to the respiratory tract. DK/Anyang/AVL-1/01 virus caused weight loss and resulted in 22 to 33% mortality, whereas CK/HK/317.5/01-infected mice exhibited no morbidity or mortality. The isolation of a highly pathogenic H5N1 influenza virus from poultry indicates that such viruses are still circulating in China and may present a risk for transmission of the virus to humans.