Online Semi-supervised Multi-label Classification with Label Compression and Local Smooth Regression

Authors: Peiyan Li, Honglian Wang, Christian Böhm, Junming Shao

IJCAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments provide empirical evidence for the effectiveness of our approach.
Researcher Affiliation Academia 1Data Mining Lab, University of Electronic Science and Technology of China 2Ludwig-Maximilians-Universit at M unchen
Pseudocode Yes Algorithm 1 Budget Maintenance for BL and BA, Algorithm 2 Adaptive Update Strategy for Q, Algorithm 3 Adjustment Strategy for Q, Algorithm 4 On Se ML
Open Source Code No The paper does not provide any explicit statements about releasing source code for the methodology described, nor does it provide a link to a code repository.
Open Datasets Yes All data sets employed in our experiments are publicly available on Mulan1 [Tsoumakas et al., 2009] and Lambda2 [Zhang and Zhou, 2010].
Dataset Splits No The paper states, "We set 10% or 20% of the instances to be labeled, and the rest to be unlabeled," and mentions random splits for experiments, but it does not explicitly specify a separate "validation" dataset split.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details, such as library names with version numbers, needed to replicate the experiment.
Experiment Setup Yes For On Se ML, the budget size s L and s A are set to bound the model size. For larger data sets, e.g. Mirflickr and Mediamill, we simply set s L = 0.05N and s A = 0.1N; while for small data sets like Enron and Corel5k, we leave the model size unbounded. For the second term in Eq. (7), the number of nearest neighbors is set to 10. The regularization parameter λ is tuned in {0.0001, 0.001, 0.1}. As for k and s Q, they are largely decided by the size of the data set. Generally, we set k = 0.1d and tuning s Q in the range of {5, 10, 20, 40}. For TRAM, BSSML and OBR, we use their default parameter settings. For OELM, we tune the number of hidden layer neurons between {100, 200, 300, 400, 500} and the batch size from 10 to 100 with the step size 10.