Paper notes: multi label learning ackel

Abstract : Share your understanding of the paper . See the original Wang, R., Kwong, S., Wang, X., & Jia, Y. (2021). Active k-labelsets ensemble for multi-label classification. Pattern Recognition, 109, 107583.

1. Contribution of thesis

Solve the problem of random attribute set selection .

2. The main idea

Random $k$-labelsets ensemble (RA$k$EL) Randomly put $k$ Tags are merged into one with $2^k$ Decision attributes with attribute values . Such as : cat + dog $\to$ { $\varnothing$, {cat}, {dog}, {cat&dog}}. Its advantage is

• It reflects the relevance of labels ;
• Attributes inside 0 value ( There are no cats or dogs here ) less , Effectively mitigate category imbalances ;
• It can be set according to the needs of users $k$;
• Multiple classifiers can be further fused .

AC$k$EL Added biased attribute selection , Avoid the drawbacks of random selection .

3. Summary

Turn multi label learning problem into multi category learning problem , This is an original idea . The specific moves in the back are all plain .

---

To be continued