Bidirectional Active Learning with Gold-Instance-Based Human Training

Authors: Feilong Tang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on three real datasets demonstrate that our BALT algorithm significantly outperforms representative related proposals.To validate our proposed BALT model and algorithm, we conducted experiments with real human on three real datasets.
Researcher Affiliation Academia Feilong Tang Department of Computer Science and Engineering, Shanghai Jiao Tong University, China tang feilong@sjtu.edu.cn
Pseudocode Yes Algorithm 1: BALT Algorithm
Open Source Code No The paper does not provide concrete access to its source code via a link, explicit statement of release, or mention of code in supplementary materials.
Open Datasets Yes Bird species classification on the Caltech-UCSD Birds dataset [Welinder et al., 2010]...Leaf species classification on the Leafsnap dataset [Kumar et al., 2012]...Butterfly species classification on the Butterflies dataset. We manually selected a total of 306 butterfly images representing 8 species from Image Net.
Dataset Splits No For each dataset, we partitioned the samples into three subsets: gold instances, unlabeled instances and testing instances, with proportions of 5%, 65% and 30%, respectively.The remaining 65% and 30% of the samples were used for training and testing, respectively. While train and test splits are mentioned, a distinct "validation" split is not provided.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No We used the basic logistic regression classifier with L2 regularization implemented in Weka in combination with all of these methods.The paper mentions "Weka" but does not specify a version number for Weka or any other software dependency.
Experiment Setup Yes In our BALT algorithm, we set the initial probability of gold instance sampling as ρ(0)=0.2 and set the following parameter weights: σ=0.1 and λ=λ1=λ2=0.05.