HAR: Hardness Aware Reweighting for Imbalanced Datasets

Duggal, Rahul; Freitas, Scott; Dhamnani, Sunny; Chau, Duen Horng; Sun, Jimeng

doi:10.1109/bigdata52589.2021.9671807

Cited by 6 publications

(8 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Note that, if we can increase the depth of our network and use more classifier branches, the model performance will be better. Moreover, the performance of MCCM dir is better than that of MCCM conf , which means that our designed sample difficulty discrimination module is better than that proposed in ELF ( Duggal et al, 2020 ). The Dirichlet distribution united with the information entropy can divide the samples into hard and easy ones, which is better than using the simple classifier confidience.…”

Section: Resultsmentioning

confidence: 86%

“…MCCM conf : the variant is designed based on the ELF ( Duggal et al, 2020 ), which uses the classifier confidence to distinguish the samples into easy and hard ones. Particularly, in the training process, if the samples are classified correctly and their classifier confidence is lower than the threshold (0.9 used in ELF and this study), they will not be sent to the hard classifier.…”

Section: Resultsmentioning

confidence: 99%

“…Predicting protein secondary structure in genetics and bioinformatics is a challenging task that we try to solve from the perspective of the model learning method. On the one hand, the existing research results point out that, within different classes of all samples (the classes are either majority or minority), some examples are easier than the others ( Duggal et al, 2020 ). On the other hand, different people may be suitable for different work.…”

Section: Methodsmentioning

confidence: 99%

“…Integrated and developed multiple advanced deep learning architectures (DNSS2) to further improve secondary structure prediction. As described above, most researchers currently focus on exploring the complex deep learning models, and a few try to solve the protein secondary structure prediction task from the perspective of model learning or training methods, for example, “ELF: An Early-Exiting Framework for Long-Tailed Classification” ( Duggal et al, 2020 ).…”

Section: Introductionmentioning

confidence: 99%

“…However, by treating each example within a class equally, these methods fail to account for the important notion of example hardness. In other words, within each class, some examples are easier to classify than others ( Duggal et al, 2020 ). Hence, “ELF: An Early-Exiting Framework for Long-Tailed Classification” is proposed to overcome the above-described limitations and address the challenge of data imbalance.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Multistage Combination Classifier Augmented Model for Protein Secondary Structure Prediction

et al. 2022

View full text Add to dashboard Cite

In the field of bioinformatics, understanding protein secondary structure is very important for exploring diseases and finding new treatments. Considering that the physical experiment-based protein secondary structure prediction methods are time-consuming and expensive, some pattern recognition and machine learning methods are proposed. However, most of the methods achieve quite similar performance, which seems to reach a model capacity bottleneck. As both model design and learning process can affect the model learning capacity, we pay attention to the latter part. To this end, a framework called Multistage Combination Classifier Augmented Model (MCCM) is proposed to solve the protein secondary structure prediction task. Specifically, first, a feature extraction module is introduced to extract features with different levels of learning difficulties. Second, multistage combination classifiers are proposed to learn decision boundaries for easy and hard samples, respectively, with the latter penalizing the loss value of the hard samples and finally improving the prediction performance of hard samples. Third, based on the Dirichlet distribution and information entropy measurement, a sample difficulty discrimination module is designed to assign samples with different learning difficulty levels to the aforementioned classifiers. The experimental results on the publicly available benchmark CB513 dataset show that our method outperforms most state-of-the-art models.

show abstract

Section: Resultsmentioning

confidence: 86%

Section: Resultsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations