Annotated Data Research Articles

An efficient zero-labeling segmentation approach for pest monitoring on smartphone-based images

Timely and precise farm inspection, which involves the identification and recognition of harmful insects and diseases, is crucial for safeguarding crop production. Traditional vision-based pest recognition methods typically require extensive annotated data for each pest species and a lengthy training process. This approach is time-consuming, labor-intensive, and prone to human error. Zero-shot learning offers a potential solution by enabling pest segmentation and control without requiring explicit training data. This study supports farmers in automatically identifying ten common pests and their precise locations in real-world outdoor environments. The zero-shot pest segmentation is based on a hybrid approach combining Explainable Contrastive Language-Image Pre-training (ECLIP) and Segment-Anything (SAM). Moreover, an optimized super-resolution model and various data augmentation methods are implemented to improve the quality of the dataset. Lastly, a mask post-processing step is applied to remove highly overlapping segmented masks and noise blobs caused by the complex background. The mean Intersection over Union (mIoU) of 66.5 % on the validation set demonstrates the potential of zero-shot methods for automated pest segmentation during farm inspections. This research lays the foundation for accurate pest monitoring systems capable of adapting to new pests, ultimately improving agricultural productivity.

Application of Generative Adversarial Networks in Semi-Annotated Defect Synthesis and Detection Under Limited Resources

Defect detection is a crucial technology that is extensively employed in the manufacturing industry to monitor and ensure the quality of output. Deep learning models have shown remarkable potential for defect detection. However, the success of these models heavily relies on voluminous training data. Collecting substantial amounts of defect data is challenging in practical settings, and the tedious process of pixel-level defect annotation further complicates the task. Among the common defects encountered in manufacturing, scratches are particularly significant. To address these challenges, this study proposes a two-phase generative adversarial network (GAN) approach for synthesizing defect images and generating semi-automatic pixel-wise labels for anomaly detection. The first phase primarily focuses on synthesizing images, while the second phase involves the pixel-wise labeling of the images. The synthesized paired images generated by the GANs serve as input to the semantic network. Notably, the proposed methodology requires only a few real defect samples for training and a small amount of annotated data, making it practical and computationally efficient for implementation in the manufacturing industry. Experimental results indicate the effectiveness of the proposed deep-learning solution in defect detection, specifically in identifying scratches on various textured and patterned surfaces. A notably high detection accuracy is achieved, validating the potential of the approach in real-world manufacturing scenarios.

A general framework for generative self-supervised learning in non-invasive estimation of physiological parameters using photoplethysmography

Aligning physiological parameter labels with large-scale photoplethysmographic (PPG) data for deep learning is challenging and resource-intensive. While self-supervised representation learning (SSRL) can handle limited annotated data, the challenge lies in learning robust shared representations from vast unlabeled data and integrating various contextual cues to learn distinctive representations. To alleviate these challenges, a generative SSRL framework TS2TC is proposed to collaboratively utilize the temporal, spectrogram, and temporal-spectrogram mixed domains to explore and incorporate the unique features of PPG for universal and non-invasive physiological parameter estimation. Initially, a pretext task named Cross-Temporal Fusion Generative Anchor (CTFGA) is designed, modeling temporal dependencies and reconstructing independent segments at a coarse level to provide robust global feature extraction and local semantic contextual representation. The framework also includes sub-signals from PPG with diverse frequency scales and order derivatives reflecting hemodynamics to facilitate learning shared representations at varying semantic levels. Secondly, an advanced cognitive-inspired dual-process transfer (DPT) strategy is formulated, consisting of prior-dependent autonomous processes and posterior observation reasoning processes, to leverage the independent and integrated advantages of shared and specific representations. Furthermore, TS2TC introduces a novel bilinear temporal-spectrogram fusion method in the mixed domain, aligning latent representations from different domains, and establishing fine-grained contextual interactions at the feature level across multiple sources of information. Extensive experiments on physiological parameter estimation tasks showed that the joint performance of CTFGA and DPT outperforms standard generative learning significantly. TS2TC achieved an average 2.49% improvement in RMSE over the current state-of-the-art estimation methods with only 10% training data.

Comprehensive scRNA-seq analysis to identify new markers of M2 macrophages for predicting the prognosis of prostate cancer

Background Prostate cancer (PCa) has become the highest incidence of malignant tumor among men in the world. Tumor microenvironment (TME) is necessary for tumor growth. M2 macrophages play an important role in many solid tumors. This research aimed at the role of M2 macrophages’ prognosis value in PCa. Methods Single-cell RNA-seq (scRNA-seq) data and mRNA expression data were obtained from the Gene Expression Omnibus database (GEO) and The Cancer Genome Atlas (TCGA). Quality control, normalization, reduction, clustering, and cell annotation of scRNA-seq data were preformed using the Seruat package. The sub-populations of the tumor-associated macrophages (TAMs) were analysis and the marker genes of M2 macrophage were selected. Differentially expressed genes (DEGs) in PCa were identified using limma and the immune infiltration was detected using CIBERSORTx. Then, a weighted correlation network analysis (WGCNA) was constructed to identify the M2 macrophage-related modules and genes. Integration of the marker genes of M2 macrophage from scRNA-seq data analysis and hub genes from WGCNA to select the prognostic gene signature based on Univariate and LASSO regression analysis. The risk score was calculated, and the DEGs, biological function, immune characteristics related to risk score were explored. And a predictive nomogram was constructed. CCK8, Transwell, and wound healing were used to verify cell phenotype changes after co-cultured. Results A total of 2431 marker genes of M2 macrophage and 650 hub M2 macrophage-related genes were selected based on scRNA-seq data and WGCNA. Then, 113 M2 macrophage-related genes were obtained by overlapping the scRNA-seq data and WGCNA results. Nine M2 macrophage-related genes (SMOC2, PLPP1, HES1, STMN1, GPR160, ABCG1, MAZ, MYC, and EPCAM) were screened as prognostic gene signatures. M2 risk score was calculated, the DEGs, Immune score, stromal score, ESTIMATE score, tumor purity, and immune cell infiltration, immune checkpoint expression, and responses of immunotherapy and chemotherapy were identified. And a predictive nomogram was constructed. CCK8, Transwell invasion, and wound healing further verified that M2 macrophages promoted the proliferation, invasion, and migration of PCa (p < 0.05). Conclusions We uncovered that M2 macrophages and relevant genes played key roles in promoting the occurrence, development, and metastases of PCa and played as convincing predictors in PCa.

High-quality expert annotations enhance artificial intelligence model accuracy for osteosarcoma X-ray diagnosis.

Primary malignant bone tumors, such as osteosarcoma, significantly affect the pediatric and young adult populations, necessitating early diagnosis for effective treatment. This study developed a high-performance artificial intelligence (AI) model to detect osteosarcoma from X-ray images using highly accurate annotated data to improve diagnostic accuracy at initial consultations. Traditional models trained on unannotated data have shown limited success, with sensitivities of approximately 60%-70%. In contrast, our model used a data-centric approach with annotations from an experienced oncologist, achieving a sensitivity of 95.52%, specificity of 96.21%, and an area under the curve of 0.989. The model was trained using 468 X-ray images from 31 osteosarcoma cases and 378 normal knee images with a strategy to maximize diversity in the training and validation sets. It was evaluated using an independent dataset of 268 osteosarcoma and 554 normal knee images to ensure generalizability. By applying the U-net architecture and advanced image processing techniques such as renormalization and affine transformations, our AI model outperforms existing models, reducing missed diagnoses and enhancing patient outcomes by facilitating earlier treatment. This study highlights the importance of high-quality training data and advocates a shift towards data-centric AI development in medical imaging. These insights can be extended to other rare cancers and diseases, underscoring the potential of AI in transforming diagnostic processes in oncology. The integration of this AI model into clinical workflows could support physicians in early osteosarcoma detection, thereby improving diagnostic accuracy and patient care.

Transformer-based Named Entity Recognition for Clinical Cancer Drug Toxicity by Positive-unlabeled Learning and KL Regularizers

Background: With increasing rates of polypharmacy, the vigilant surveillance of clinical drug toxicity has emerged as an important With increasing rates of polypharmacy, the vigilant surveillance of clinical drug toxicity has emerged as an important concern. Named Entity Recognition (NER) stands as an indispensable undertaking, essential for the extraction of valuable insights regarding drug safety from the biomedical literature. In recent years, significant advancements have been achieved in the deep learning models on NER tasks. Nonetheless, the effectiveness of these NER techniques relies on the availability of substantial volumes of annotated data, which is labor-intensive and inefficient. Methods: This study introduces a novel approach that diverges from the conventional reliance on manually annotated data. It employs a transformer-based technique known as Positive-Unlabeled Learning (PULearning), which incorporates adaptive learning and is applied to the clinical cancer drug toxicity corpus. To improve the precision of prediction, we employ relative position embeddings within the transformer encoder. Additionally, we formulate a composite loss function that integrates two Kullback-Leibler (KL) regularizers to align with PULearning assumptions. The outcomes demonstrate that our approach attains the targeted performance for NER tasks, solely relying on unlabeled data and named entity dictionaries. Conclusion: Our model achieves an overall NER performance with an F1 of 0.819. Specifically, it attains F1 of 0.841, 0.801 and 0.815 for DRUG, CANCER, and TOXI entities, respectively. A comprehensive analysis of the results validates the effectiveness of our approach in comparison to existing PULearning methods on biomedical NER tasks. Additionally, a visualization of the associations among three identified entities is provided, offering a valuable reference for querying their interrelationships.

Fast clustering and cell-type annotation of scATAC data using pre-trained embeddings.

Data from the single-cell assay for transposase-accessible chromatin using sequencing (scATAC-seq) are now widely available. One major computational challenge is dealing with high dimensionality and inherent sparsity, which is typically addressed by producing lower dimensional representations of single cells for downstream clustering tasks. Current approaches produce such individual cell embeddings directly through a one-step learning process. Here, we propose an alternative approach by building embedding models pre-trained on reference data. We argue that this provides a more flexible analysis workflow that also has computational performance advantages through transfer learning. We implemented our approach in scEmbed, an unsupervised machine-learning framework that learns low-dimensional embeddings of genomic regulatory regions to represent and analyze scATAC-seq data. scEmbed performs well in terms of clustering ability and has the key advantage of learning patterns of region co-occurrence that can be transferred to other, unseen datasets. Moreover, models pre-trained on reference data can be exploited to build fast and accurate cell-type annotation systems without the need for other data modalities. scEmbed is implemented in Python and it is available to download from GitHub. We also make our pre-trained models available on huggingface for public use. scEmbed is open source and available at https://github.com/databio/geniml. Pre-trained models from this work can be obtained on huggingface: https://huggingface.co/databio.

Predicting cell cycle stage from 3D single-cell nuclear-stained images.

The cell cycle governs the proliferation, differentiation, and regeneration of all eukaryotic cells. Profiling cell cycle dynamics is therefore central to basic and biomedical research spanning development, health, aging, and disease. However, current approaches to cell cycle profiling involve complex interventions that may confound experimental interpretation. To facilitate more efficient cell cycle annotation of microscopy data, we developed CellCycleNet, a machine learning (ML) workflow designed to simplify cell cycle staging with minimal experimenter intervention and cost. CellCycleNet accurately predicts cell cycle phase using only a fluorescent nuclear stain (DAPI) in fixed interphase cells. Using the Fucci2a cell cycle reporter system as ground truth, we collected two benchmarking image datasets and trained two ML models-a support vector machine (SVM) and a deep neural network-to classify nuclei as being in either the G1 or S/G2 phases of the cell cycle. Our results suggest that CellCycleNet outperforms state-of-the-art SVM models on each dataset individually. When trained on two image datasets simultaneously, CellCycleNet achieves the highest classification accuracy, with an improvement in AUROC of 0.08-0.09. The model also demonstrates excellent generalization across different microscopes, achieving an AUROC of 0.95. Overall, using features derived from 3D images, rather than 2D projections of those same images, significantly improves classification performance. We have released our image data, trained models, and software as a community resource.

Development and application of the GenoBaits WheatSNP16K array to accelerate wheat genetic research and breeding

Single-nucleotide polymorphisms (SNPs) are widely used as molecular markers for constructing genetic linkage maps in wheat. Compared with available SNP-based genotyping platforms, a genotyping by target sequencing (GBTS) system with capture-in-solution (liquid chip) technology has become the favored genotyping technology because it is less demanding and more cost effective, flexible, and user-friendly. In this study, a new GenoBaits WheatSNP16K (GBW16K) GBTS array was designed using datasets generated by the wheat 660K SNP array and resequencing platforms in our previous studies. The GBW16K array contains 14 868 target SNP regions that are evenly distributed across the wheat genome, and 37 669 SNPs in these regions can be identified in a diversity panel consisting of 239 wheat accessions from around the world. Principal component and neighbor-joining analyses using the called SNPs are consistent with the pedigree information and geographic distributions or ecological environments of the accessions. For the GBW16K marker panel, the average genetic diversity among the 239 accessions is 0.270, which is sufficient for linkage map construction and preliminary mapping of targeted genes or quantitative trait loci (QTLs). A genetic linkage map, constructed using the GBW16K array-based genotyping of a recombinant inbred line population derived from a cross of the CIMMYT wheat line Yaco“S” and the Chinese landrace Mingxian169, enables the identification of Yr27, Yr30, and QYr.nwafu-2BL.4 for adult-plant resistance to stripe rust from Yaco“S” and of Yr18 from Mingxian169. QYr.nwafu-2BL.4 is different from any previously reported gene/QTL. Three haplotypes and six candidate genes have been identified for QYr.nwafu-2BL.4 on the basis of haplotype analysis, micro-collinearity, gene annotation, RNA sequencing, and SNP data. This array provides a new tool for wheat genetic analysis and breeding studies and for achieving durable control of wheat stripe rust.

A Generalisable Heartbeat Classifier Leveraging Self-Supervised Learning for ECG Analysis During Magnetic Resonance Imaging.

Electrocardiogram (ECG) is acquired during Magnetic Resonance Imaging (MRI) to monitor patients and synchronize image acquisition with the heart motion. ECG signals are highly distorted during MRI due to the complex electromagnetic environment. Automated ECG analysis is therefore complicated in this context and there is no reference technique in MRI to classify pathological heartbeats. Imaging arrhythmic patients is hence difficult in MRI. Deep Learning based heartbeat classifier have been suggested but require large databases whereas existing annotated sets of ECG in MRI are very small. We proposed a Siamese network to leverage a large database of unannotated ECGs outside MRI. This was used to develop an efficient representation of ECG signals, further used to develop a heartbeat classifier. We extensively tested several data augmentations and self-supervised learning (SSL) techniques and assessed the generalization of the obtained classifier to ECG signals acquired in MRI. These augmentations included random noises and a model simulating MRI specific artefacts. SSL pretraining improved the generalizability of heartbeat classifiers in MRI (F1=0.75) compared to Deep Learning not relying on SSL (F1=0.46) and another classical machine learning approach (F1=0.40). These promising results seem to indicate that the use of SSL techniques can learn efficient ECG signal representation, and are useful for the development of Deep Learning models even when only scarce annotated medical data are available.

Few-Shot Object Detection: A Comprehensive Survey.

Humans are able to learn to recognize new objects even from a few examples. In contrast, training deep-learning-based object detectors requires huge amounts of annotated data. To avoid the need to acquire and annotate these huge amounts of data, few-shot object detection (FSOD) aims to learn from few object instances of new categories in the target domain. In this survey, we provide an overview of the state of the art in FSOD. We categorize approaches according to their training scheme and architectural layout. For each type of approach, we describe the general realization as well as concepts to improve the performance on novel categories. Whenever appropriate, we give short takeaways regarding these concepts in order to highlight the best ideas. Eventually, we introduce commonly used datasets and their evaluation protocols and analyze the reported benchmark results. As a result, we emphasize common challenges in evaluation and identify the most promising current trends in this emerging field of FSOD.

Enhancing training performance of convolutional neural network algorithm through an autoencoder-based unsupervised labeling framework for mineral exploration targeting

Convolutional neural networks (CNNs), a prominent deep learning approach, have garnered significant interest in the field of mineral potential mapping (MPM) due to their capability to capture and learn spatial features that traditional algorithms tend to overlook. The effectiveness of CNNs is closely tied to the quantity of training data available, thereby impacting the outcomes of MPM. Moreover, uncertainties arising from delineation of negative samples can compromise the reliability of MPM assessments. To deal with these challenges, we propose the utilization of an autoencoder-based anomaly detection technique for the purpose of annotating locations in an unsupervised manner. Subsequently, a CNN is trained using the unsupervised annotated data to generate a Fe prospectivity model within a specific region in Iran. To validate the effectiveness of the proposed method, we first perform an MPM with insufficient positive training samples that extract from the location of known occurrences. We then execute another MPM on a set of samples labeled based on the reconstruction error of an autoencoder network. When comparing the two prospectivity models, namely using augmented data or inadequate samples, it is evident that modeling with augmented data outperforms the MPM model trained with insufficient samples. This confirms the effectiveness of the adopted approach and shows that the unsupervised labeling technique proposed in this work can significantly improve the performance of the CNN in MPM.

PIS-Net: Efficient weakly supervised instance segmentation network based on annotated points for rice field weed identification

Weed damage in rice fields is one of the main causes of reduced rice yields and quality. Accurate and efficient weed identification is the prerequisite for realizing intelligent and precise weeding in paddies. Recently, Vision Transformers (ViTs) have emerged with superior performance on computer vision tasks compared to the convolutional neural network (CNN)-based models. However, the lack of fully labeled weed datasets hinders the potential application of deep learning models in weed identification. To address the above issues, this study customizes a novel point-supervised instance segmentation network (PIS-Net) for weakly supervised instance segmentation of weeds in rice fields. More correctly, we first propose a novel instance segmentation point labeling scheme that utilizes randomly generated annotation points within each instance, aiming to decrease both labeling time and difficulty. Additionally, to make optimal use of point labels, this study puts forth a mask generation strategy based on the adaptive selection of pyramid levels. In this sense, the network model can flexibly choose the pyramid level expected to generate the most suitable instance mask based on the network's reliability. Finally, we establish the pseudo label refinement network (PLR-Net) to refine rough instance masks. The proposed PIS-Net utilizes 13 randomly generated annotation points for each instance, yet achieving an AP of 38.5 and an AP50 of 68.3, which is superior to the baseline mask-R-CNN with an AP of 8.2 and AP50 of 6.9, achieving 90 % fully supervised performance. This method effectively utilizes point labels, annotated with high efficiency, as a robust source of weak supervision to address challenges in weed data annotation and the low accuracy of existing weakly supervised models. Experiments show that the point annotation scheme of the PIS-Net is faster than full-object mask annotation, and the AP is also higher than the current semi-supervised weed segmentation model, enjoying high potentials in practical paddy fields.

Highly contiguous genome assembly and gene annotation of the short-finned eel (Anguilla bicolor pacifica)

In East Asia, anguillid eels are commercially important. However, unlike other species, they have not been successfully cultivated throughout their lifecycle. Facing population decline due to overharvesting and environmental pressures, the industry is turning to alternatives, such as Anguilla bicolor pacifica (short-finned eel). However, genomic data for short-finned eels are unavailable. Here, we present in-depth whole-genome sequencing results for short-finned eel obtained using two sequencing platforms (PacBio Revio, and Illumina). In this study, we achieved a highly contiguous genome assembly of the short-finned eel, comprising 19 pseudochromosomes encompassing 99.76% of the 1.087 Gb genome sequence with an N50 of 16.88 and 61.07 Mb from contig and scaffold, respectively. Transcripts from four different tissues led to the annotation of 23,095 protein-coding genes in the eel genome, 98.66% of which were functionally annotated. This high-quality genome assembly, along with the annotation data, provides a foundation for future functional genomic studies of short-finned eels.

Transcriptome Analysis of Multiple Plant Parts in the Woody Oil Tree Camellia drupifera Loureiro

Camellia drupifera is mainly used in forestry for its high-value industrial products; however, limited information is available on its transcriptome. This study aimed to construct a full-length transcriptome sequence based on the PacBio sequencing platform for various plant parts of C. drupifera, including flower buds, leaves, leaf buds, branches, the pericarp, and seed kernels. The transcriptomes were annotated with 23,207 genes, with 58 subgroups in the GO classification. The KEGG database revealed 10,407 genes involved in the metabolic pathway analysis, with 68,192 coding sequences, 3352 TF families, 48,541 SSRs, 1421 IncRNAs, and 2625 variable shears predicted. The transcriptomes of different parts were analyzed and compared. The majority of differentially expressed genes (DEGs) were found between the pericarp and seed kernels, followed by leaves and the pericarp with 5662 DEGs, and flower buds and leaf buds with 1616 DEGs. GO and KEGG enrichment analyses showed that KEGG differential genes were significant in microbial metabolism, carbon metabolism, and other functions. The data annotation and analysis of the full-length transcriptome and the comparative analysis between different plant parts provided a theoretical basis for studying gene function, metabolic pathway regulation, and gene expression analysis in KEGG.

Identifying Legal, BIM Data and Visualization Requirements to Form Legal Spaces and Developing a Web-Based 3D Cadastre Prototype: A Case Study of Condominium Building

Over the past decade, numerous countries and researchers have been investigating the potential of 3D cadastre based on the Building Information Model (BIM). In Türkiye, the General Directorate of Land Registry and Cadastre (GDLRC) has been engaged in the “Production of 3D City Models and Creation of 3D Cadastral Bases Project” since 2018. One of the objectives is to develop 3D (physical) digital building models (and databases) through the digitization of floor plans of architectural drawings. In order to ensure the long-term viability of this project, a new regulation was issued in 2021. This regulation mandates the submission of 3D digital building models to the land registry in order to register condominiums. The future goals of the GDLRC include utilizing the 3D digital building models produced to create legal spaces in condominium buildings for the 3D cadastre. However, no research has yet been carried out for this goal. The objective of this research is to identify the legal and BIM data requirements for deriving legal spaces in condominium buildings, in light of the legislative analysis, and to develop a web-based 3D cadastre visualization prototype (showing both legal spaces and physical objects) based on the requirements obtained from the scientific literature. The result of this research demonstrates that well-structured and annotated BIM data can be used to develop a 3D cadastre prototype that meets the legal requirements in the case of Türkiye. Moreover, it is evident that, although the GDLRC has initiated the construction of a robust foundation for a 3D cadastre based on BIM, further enhancements and resolutions must be implemented from a legal and technical standpoint. It is postulated that the identified requirements and the proposed methodology in this research may assist decision-makers in Türkiye and globally in formulating their strategic plans for a 3D cadastre.

Enhanced Crime Prediction with Computer Vision-Yolov4 Approach

This research paper presents the development of an artificial intelligence safety application on an HP Pavilion gaming machine, utilizing criminal footage from reputable databases like the UCF-Crime open-source dataset. The system underwent meticulous data annotation to identify five distinct classes crucial for anomaly detection: Person, Short Gun, Handgun, Knife, and Rifle. Supervised machine learning techniques were applied, focusing on monitoring human trajectories and employing deep-SORT and Euclidean distance computations to track individuals, simulating real-world crime scenarios. The AI safety model showcased outstanding performance with an average precision rate of approximately 86.43%, exceeding 90% after 2000 iterations, demonstrating versatility across all categories with notable average precision accuracies for rifles (98.90%), handguns (96.93%), and knives (97.66%). Enhancements to the Python script improved the system's ability to detect weapons sub-objects in human subjects and classify potential perpetrators as high risk, a novel aspect of this study.

One-Shot Learning from Prototype Stock Keeping Unit Images

This paper highlights the importance of one-shot learning from prototype Stock Keeping Unit (SKU) images for efficient product recognition in retail and inventory management. Traditional methods require large supervised datasets to train deep neural networks, which can be costly and impractical. One-shot learning techniques mitigate this issue by enabling classification from a single prototype image per product class, thus reducing data annotation efforts. We introduce the Variational Prototyping Encoder (VPE), a novel deep neural network for one-shot classification. Utilizing a support set of prototype SKU images, VPE learns to classify query images by capturing image similarity and prototypical concepts. Unlike metric learning-based approaches, VPE pre-learns image translation from real-world object images to prototype images as a meta-task, facilitating efficient one-shot classification with minimal supervision. Our research demonstrates that VPE effectively reduces the need for extensive datasets by utilizing a single image per class while accurately classifying query images into their respective categories, thus providing a practical solution for product classification tasks.

SPORT: A Subgraph Perspective on Graph Classification with Label Noise

Graph neural networks (GNNs) have achieved great success recently on graph classification tasks using supervised end-to-end training. Unfortunately, extensive noisy graph labels could exist in the real world because of the complicated processes of manual graph data annotations, which may significantly degrade the performance of GNNs. Therefore, we investigate the problem of graph classification with label noise, which is demanding because of the complex graph representation learning issue and serious memorization of noisy samples. In this work, we present a novel approach called S ubgra p h Set Netw or k with Sample Selection and Consis t ency Learning (SPORT) for this problem. To release the overfitting of GNNs, SPORT proposes to characterize each graph as a set of subgraphs generated by certain predefined stratagems, which can be viewed as samples from its underlying semantic distribution in graph space. Then we develop an equivariant network to encode the subgraph set with the consideration of the symmetry group. To further release the influences of noisy examples, we leverage the predictions of subgraphs to measure the likelihood of a sample being clean or noisy, followed by effective label updating. In addition, we propose a joint loss to advance the model generalizability by introducing consistency regularization. Comprehensive experiments on a wide range of graph classification datasets demonstrate the effectiveness of our SPORT. Specifically, SPORT outperforms the most competing baseline by up to 6.4%.

Towards Cross-lingual Social Event Detection with Hybrid Knowledge Distillation

Recently published graph neural networks (GNNs) show promising performance at social event detection tasks. However, most studies are oriented toward monolingual data in languages with abundant training samples. This has left the common lesser-spoken languages relatively unexplored. Thus, in this work, we present a GNN-based framework that integrates cross-lingual word embeddings into the process of graph knowledge distillation for detecting events in low-resource language data streams. To achieve this, a novel cross-lingual knowledge distillation framework, called CLKD, exploits prior knowledge learned from similar threads in English to make up for the paucity of annotated data. Specifically, to extract sufficient useful knowledge, we propose a hybrid distillation method that consists of both feature-wise and relation-wise information. To transfer both kinds of knowledge in an effective way, we add a cross-lingual module in the feature-wise distillation to eliminate the language gap and selectively choose beneficial relations in the relation-wise distillation to avoid distraction caused by teachers’ misjudgments. Our proposed CLKD framework also adopts different configurations to suit both offline and online situations. Experiments on real-world datasets show that the framework is highly effective at detection in languages where training samples are scarce.