Enhancing Multi-Label Classification via Dynamic Label-Order Learning

Authors: Jiangnan Li, Yice Zhang, Shiwei Chen, Ruifeng Xu

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on public datasets reveal that our approach greatly outperforms previous methods.
Researcher Affiliation Academia 1Harbin Institute of Technology, Shenzhen, China 2Peng Cheng Laboratory, Shenzhen, China 3Guangdong Provincial Key Laboratory of Novel Security Intelligence Technologies
Pseudocode Yes Algorithm 1: The proposed label-order learning algorithm
Open Source Code Yes We will release our code at https: //github.com/Kagami Baka/DLOL.
Open Datasets Yes The dataset used is Reuters-21578 (Hayes and Weinstein 1990). and we evaluate our approach on four typical multilabel classification datasets, namely Reuters-21578, RCV1-V2, Slashdot, and Go Emotions.
Dataset Splits Yes Table 3: Statistics of four datasets. (includes columns for Train, Dev, Test sample counts)
Hardware Specification No The paper does not provide specific details about the hardware used for experiments, such as GPU or CPU models.
Software Dependencies No The paper mentions using BART-base as the backbone model but does not specify software dependencies with version numbers (e.g., Python, PyTorch, or CUDA versions).
Experiment Setup Yes Our approach has three hyper-parameters for multi-reference training, label smoothing, and eos penalty... the optimal values for these hyper-parameters are α = 0.1, β = 0.1, and γ = 0.9, respectively.