Deep Streaming Label Learning (DSLL)

• Exploring and exploiting the knowledge from past labels and historical models to understand and develop emerging new labels.

• Framework: PyTorch

Dependence

• python 3.7
• Pytoch 1.4
• liac-arff (pip install liac-arff)

Running

cd ./ # open the folder where the code is located

python3 DSLL.py # run DSLL model

Results

Figure 1. Performance comparison of learning new labels with different batch sizes by considering 50% of labels as past labels. m indicates the number of new labels.

Table 1. Ranking performance of each comparison algorithm for learning new labels with different batch sizes by regarding 50% of labels as past labels. #label denotes the number of new labels. means the smaller (larger) the value, the better the performance.

More detailed results can be found in the Supplementary Materials.

Cite

@inproceedings{icml2020_230,
Author = {Wang, Zhen and Liu, Liu and Tao, Dacheng},
Booktitle = {International Conference on Machine Learning (ICML)},
Pages = {378--387},
Title = {Deep Streaming Label Learning},
Year = {2020}}