Partial Multi-Label Learning by Low-Rank and Sparse Decomposition

Authors: Lijuan Sun, Songhe Feng, Tao Wang, Congyan Lang, Yi Jin5016-5023

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments In this section, we first describe our experimental setup, including the benchmark data sets, comparing algorithms, and evaluation metrics. Then we present three sets of experiments to verify the effectiveness of the proposed PML-LRS approach...
Researcher Affiliation Academia Lijuan Sun, Songhe Feng, Tao Wang, Congyan Lang, Yi Jin School of Computer and Information Technology, Beijing Jiaotong University {17112082, shfeng, twang, cylang, yjin}@bjtu.edu.cn
Pseudocode Yes The optimization problem (7) can be solved with the ALM (Zhang et al. 2017), which minimizes the following augmented Lagrange function: ... Eq. (9) can be solved iteratively via the following subproblems: 1. When keeping P, Q, J, T fixed, we obtain the following equation for W by taking the derivative of Eq. (9), denoted by LRS-1... 2. When P, Q, W, T are fixed, optimizing Eq. (9) with respect to J is equivalent to the following problem, denoted by LRS-2... 3. Fixing W, J, T, solve (9) for P and Q by the following problem, denoted by LRS-3... 4. With P, Q, W, J fixed, the computation of T is independent, we obtain the following optimization problem for T by taking the derivative of Eq. (9), denoted by LRS-4...
Open Source Code No The paper does not provide concrete access to source code for the methodology described.
Open Datasets Yes We perform experiments on six data sets. These data sets spanned a broad range of applications: corel5k for image annotation, CAL500 and emotions for music classification, genbase for protein classification, medical for text categorization and delicious for web categorization. Ten-fold cross-validation is performed on the benchmark data sets... followed by Table 1 with specific dataset names and citations like 'emotions (Trohidis et al.2008)'.
Dataset Splits Yes Ten-fold cross-validation is performed on the benchmark data sets, where the mean metric value, as well as standard deviation, are recorded for each comparing algorithm.
Hardware Specification No The paper does not provide specific hardware details used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers.
Experiment Setup Yes At last, we study the influences of the three parameters, γ, β and η for the proposed method on the medical data set. Our experiment is accomplished by using the grid search method which conducts the parameter analysis by varying three parameters simultaneously. The experimental results are shown in Figure 3 which are measured by the five evaluation metrics. It can be seen that how the performance of our algorithm varies as these parameters change. Therefore we should safely set them in a wide range in practice. From this figure, we can notice that better performances are gained when γ = 0.01, β = 0.1, and η = 1.