Synthetic Oversampling of Multi-label Data Based on Local Label Distribution

Liu, Bin; Tsoumakas, Grigorios

doi:10.1007/978-3-030-46147-8_11

Cited by 18 publications

(10 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The concept of local imbalance concerns the label distribution of an instance within the local region, playing a key role in determining the difficulty of a dataset to be learned [21]. Concerning DTI data that contain two kinds of similarities, the local imbalance can be assessed in both drug space and target space.…”

Section: Local Imbalancementioning

confidence: 99%

“…It updates the original interaction matrix based on the neighborhood information to mitigate the impact of missing interactions. In addition, WkNNIR exploits the concept of local class imbalance [21] to weigh drug and target similarities, which boosts interaction recovery and prediction. Furthermore, we propose three ensemble methods to further improve the performance of WkNNIR and other DTI prediction methods.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Drug-target interaction prediction via an ensemble of weighted nearest neighbors with interaction recovery

et al. 2021

Self Cite

View full text Add to dashboard Cite

Predicting drug-target interactions (DTI) via reliable computational methods is an effective and efficient way to mitigate the enormous costs and time of the drug discovery process. Structure-based drug similarities and sequence-based target protein similarities are the commonly used information for DTI prediction. Among numerous computational methods, neighborhoodbased chemogenomic approaches that leverage drug and target similarities to perform predictions directly are simple but promising ones. However, most existing similarity-based methods follow the transductive setting. These methods cannot directly generalize to unseen data because they should be re-built to predict the interactions for new arriving drugs, targets, or drug-target pairs. Besides, many similarity-based methods, especially neighborhood-based ones, cannot handle directly all three types of interaction prediction involving new drugs and/or new targets. Furthermore, a large amount of missing (undetected) interactions in current DTI datasets hinders most DTI prediction methods. To address these issues, we propose a new method denoted as Weighted k-Nearest Neighbor with Interaction Recovery (WkNNIR). Not only can WkNNIR estimate interactions of any new drugs and/or new targets

show abstract

Section: Local Imbalancementioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Drug-target interaction prediction via an ensemble of weighted nearest neighbors with interaction recovery

et al. 2021

Self Cite

View full text Add to dashboard Cite

show abstract

“…First, we adopt binary classifiers to boost the performance of the classification model. Previous works formulating this tag-recommendation problem as a multi-class multi-label classification task, but it always leads to data imbalance issues [53] as the occurrence number of each tag differs significantly. Instead, according to our observation, most related (frequent) tags attached to the UI design are exclusive with each other, such as "black vs red", "login page vs checkout page", "music app vs sport app".…”

Section: Tagging Predictionmentioning

confidence: 99%

From Lost to Found

Chen

Feng

Liu

et al. 2020

Proc. ACM Hum.-Comput. Interact.

View full text Add to dashboard Cite

Design sharing sites provide UI designers with a platform to share their works and also an opportunity to get inspiration from others' designs. To facilitate management and search of millions of UI design images, many design sharing sites adopt collaborative tagging systems by distributing the work of categorization to the community. However, designers often do not know how to properly tag one design image with compact textual description, resulting in unclear, incomplete, and inconsistent tags for uploaded examples which impede retrieval, according to our empirical study and interview with four professional designers. Based on a deep neural network, we introduce a novel approach for encoding both the visual and textual information to recover the missing tags for existing UI examples so that they can be more easily found by text queries. We achieve 82.72% accuracy in the tag prediction. Through a simulation test of 5 queries, our system on average returns hundreds more results than the default Dribbble search, leading to better relatedness, diversity and satisfaction. CCS Concepts: • Human-centered computing → User interface design; User interface management systems; Graphical user interfaces.

show abstract

“…MLSMOTE is shown to outperform the other resampling techniques. Liu and Tsoumakas [18] improve the SMOTE resampling technique by considering the local distributions of labels when determining the minority class as opposed to the global distribution of labels. This algorithm, called MLSOL, shows promising results although it has yet to be applied to videos.…”

Section: Related Workmentioning

confidence: 99%

Simultaneous and Spatiotemporal Detection of Different Levels of Activity in Multidimensional Data

et al. 2020

View full text Add to dashboard Cite

In this work, we present a novel and promising approach to autonomously detect different levels of simultaneous and spatiotemporal activity in multidimensional data. We introduce a new multilabeling technique, which assigns different labels to different regions of interest in the data, and thus, incorporates the spatial aspect. Each label is built to describe the level of activity/motion to be monitored in the spatial location that it represents, in contrast to existing approaches providing only a binary result as the presence or absence of activity. This novel Spatially and Motion-Level Descriptive (SMLD) labeling schema is combined with a Convolutional Long Short Term Memory-based network for classification to capture different levels of activity both spatially and temporally without the use of any foreground or object detection. The proposed approach can be applied to various types of spatiotemporal data captured for completely different application domains. In this paper, it was evaluated on video data as well as respiratory sound data. Metrics commonly associated with multilabeling, namely Hamming Loss and Subset Accuracy, as well as confusion matrix-based measurements are used to evaluate performance. Promising testing results are achieved with an overall Hamming Loss for video datasets close to 0.05, Subset Accuracy close to 80% and confusion matrix-based metrics above 0.9. In addition, our proposed approach's ability in detecting frequent motion patterns based on predicted spatiotemporal activity levels is discussed. Encouraging results have been obtained on the respiratory sound dataset as well, while detecting abnormalities in different parts of the lungs. The experimental results demonstrate that the proposed approach can be applied to various types of spatiotemporal data captured for different application domains.

show abstract

Synthetic Oversampling of Multi-label Data Based on Local Label Distribution

Cited by 18 publications

References 26 publications

Drug-target interaction prediction via an ensemble of weighted nearest neighbors with interaction recovery

Drug-target interaction prediction via an ensemble of weighted nearest neighbors with interaction recovery

From Lost to Found

Simultaneous and Spatiotemporal Detection of Different Levels of Activity in Multidimensional Data

Contact Info

Product

Resources

About