Multi-Positive and Unlabeled Learning

Xu, Yixing; Xu, Chang; Xu, Chao; Tao, Dacheng

doi:10.24963/ijcai.2017/444

Cited by 52 publications

(59 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Elkan et al [17] showed that the probability output by a classifier trained on positive and unlabeled examples are different. MPU [18] success to select the multiclass data from the unlabeled data by multi-positive loss. Xu et al [19] used the PU classifier to find data from unlabeled samples in the cloud based on a few labeled samples, and realized knowledge transfer using only a small number of labeled samples.…”

Section: B Positive-unlabel Classificationmentioning

confidence: 99%

Knowledge Distillation Based on Positive-Unlabeled Classification and Attention Mechanism

Tang

Liu

Jiang

et al. 2021

2021 IEEE International Symposium on Circuits and Systems (ISCAS)

View full text Add to dashboard Cite

With the rapid development of deep learning, convolutional neural networks(CNNs) have achieved great success. But these high-capability CNNs often with a huge burden of computation and memory, which hinders these CNNs from applying to practical application. To solve this problem, in this paper, we proposed a method to train a compact model with high-capacity. The student network with fewer parameters and calculations will learning from the knowledge of the teacher network with more parameters and calculations. To promote the ability of the student network, the more expressive knowledge is extracted from the middle-layer feature of neural networks by attention mechanism, and the knowledge transforms more effective from the teacher network to the student network by the positiveunlabeled(PU) classifier. We validate our method in extensive experiments, showing that it can train the student network to achieve significant performance superior to the teacher network.

show abstract

Section: B Positive-unlabel Classificationmentioning

confidence: 99%

Knowledge Distillation Based on Positive-Unlabeled Classification and Attention Mechanism

Tang

Liu

Jiang

et al. 2021

2021 IEEE International Symposium on Circuits and Systems (ISCAS)

View full text Add to dashboard Cite

show abstract

“…The negative cases (N) identified by CSRs are indeed hard cases, since their prediction scores are above the preset confidence threshold yet they are mis-classified by the existing model. It is an active research area to create classifiers with only P and U (Elkan and Noto, 2008;Xu et al, 2017). Some research has explored models that also include N, but they have been only concerned with binary classifiers (Fei and Liu, 2015;Hsieh et al, 2019;Li et al, 2010).…”

Section: Figure 1: Intelligent Customer Support Loopmentioning

confidence: 99%

A Semi-supervised Multi-task Learning Approach to Classify Customer Contact Intents

Dong¹,

Spencer²,

Biagi³

2021

Proceedings of the 4th Workshop on E-Commerce and NLP

View full text Add to dashboard Cite

In the area of customer support, understanding customers' intents is a crucial step. Machine learning plays a vital role in this type of intent classification. In reality, it is typical to collect confirmation from customer support representatives (CSRs) regarding the intent prediction, though it can unnecessarily incur prohibitive cost to ask CSRs to assign existing or new intents to the mis-classified cases. Apart from the confirmed cases with and without intent labels, there can be a number of cases with no human curation. This data composition (Positives + Unlabeled + multiclass Negatives) creates unique challenges for model development. In response to that, we propose a semi-supervised multi-task learning paradigm. In this manuscript, we share our experience in building text-based intent classification models for a customer support service on an E-commerce website. We improve the performance significantly by evolving the model from multiclass classification to semi-supervised multi-task learning by leveraging the negative cases, domain-and taskadaptively pretrained ALBERT on customer contact texts, and a number of un-curated data with no labels. In the evaluation, the final model boosts the average AUC ROC by almost 20 points compared to the baseline finetuned multiclass classification ALBERT model.

show abstract

“…Note that in addition to the above-mentioned methods, there are some other PU learning algorithms developed in recent years, such as generative adversarial PU learning [32], multi PU learning [33], semi-supervised classification-based PU learning [34], and margin-based PU learning [35].…”

Section: Positive and Unlabeled Learningmentioning

confidence: 99%

A Two-Step Classification Method Based on Collaborative Representation for Positive and Unlabeled Learning

Wang

Peng

et al. 2021

Neural Process Lett

View full text Add to dashboard Cite

Positive and Unlabeled learning (PU learning) has drawn plenty of attention among researchers over the last few years, where only labeled positive examples and unlabeled examples are available for training a classifier. Many classic techniques for solving PU learning problems belong to the category of "two-step strategy". However, quite a number of them cannot extract reliable negative examples accurately and often lead to unsatisfactory classification results. In this paper, we propose a two-step learning scheme based on the collaborative representation (CR) for PU learning. In the first step, to handle the deficiency of negative training data, collaborative representation (CR) technique is utilized to identify reliable negative examples from unlabeled training examples. Subsequently, collaborative representation based classification (CRC) framework with l 2 -norm regularization term is applied to perform PU classification. Extensive experiments on both benchmark and real-world datasets were conducted to verify the effectiveness of the proposed method, and the results demonstrate that the two-step CR-based approaches can achieve competitive classification accuracy when compared with both traditional and state-of-the-art techniques in dealing with different PU learning issues.

show abstract

Multi-Positive and Unlabeled Learning

Cited by 52 publications

References 11 publications

Knowledge Distillation Based on Positive-Unlabeled Classification and Attention Mechanism

Knowledge Distillation Based on Positive-Unlabeled Classification and Attention Mechanism

A Semi-supervised Multi-task Learning Approach to Classify Customer Contact Intents

A Two-Step Classification Method Based on Collaborative Representation for Positive and Unlabeled Learning

Contact Info

Product

Resources

About