Robust Semi-Supervised Learning through Label Aggregation

Authors: Yan Yan, Zhongwen Xu, Ivor Tsang, Guodong Long, Yi Yang

AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on five benchmark datasets demonstrate the superior effectiveness, efficiency and robustness of the proposed algorithm. In this section, we demonstrate the robustness and performance of the proposed algorithm by comparing with eight baselines.
Researcher Affiliation Academia Yan Yan, Zhongwen Xu, Ivor W. Tsang, Guodong Long, Yi Yang Centre for Quantum Computation and Intelligent Systems, University of Technology Sydney, Australia {yan.yan-3, zhongwen.xu}@student.uts.edu.au,{ivor.tsang, guodong.long, yi.yang}@uts.edu.au
Pseudocode Yes Algorithm 1 RObust Semi-Supervised Ensemble Learning (ROSSEL)
Open Source Code No The paper does not provide a link or explicit statement about releasing the source code for the described methodology.
Open Datasets Yes Five UCI datasets including CNAE9, dna, connect4, protein and rcv1 are used in the experiments.
Dataset Splits No The paper specifies a 'labeled set (5% samples), unlabeled set (75% samples) and test set (20%)' but does not explicitly define a separate validation set for hyperparameter tuning or early stopping, nor does it mention cross-validation for this purpose.
Hardware Specification Yes All experiments are conducted on a workstation with an Intel(R) CPU (Xeon(R) E5-2687W v2 @ 3.40GHz) and 32 GB memory.
Software Dependencies No The paper mentions software like LIBLINEAR and LIBSVM, but does not specify their version numbers or any other software dependencies with specific version information.
Experiment Setup Yes We select from the range of {10 5, 10 3, 10 1, 100, 101, 103, 105} for the parameters to be tuned in all methods. We empirically set the parameter k-nearest neighbour as 5 for the graph-based methods, Lap SVM and Lap RLS. We use Gaussian kernel K(xi, xj) = exp( ||xi xj||2 2σ2 ) to compute the kernel matrix. The kernel parameter σ is fixed as 1, and all feature matrices are normalized before the experiment. When sampling, we bootstrap 50% labeled data into a bootstrap replicate.