Robustness of Bayesian Pool-Based Active Learning Against Prior Misspecification

Authors: Nguyen Cuong, Nan Ye, Wee Lee

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

Reproducibility Variable Result LLM Response
Research Type Experimental We report experimental results with different priors and the mixture prior. We use the logistic regression model with different L2 regularizers... We run maximum Gibbs error with 1/σ2 = 0.01, 0.1, 0.2, 1, 10 on tasks from the 20 Newsgroups and UCI data sets (Joachims 1996; Bache and Lichman 2013) shown in the first column of Table 1. Figure 1 shows the average areas under the accuracy curves (AUC) on the first 150 selected examples for the different regularizers.
Researcher Affiliation Academia 1Department of Mechanical Engineering, National University of Singapore, Singapore, nvcuong@nus.edu.sg 2Mathematical Sciences School & ACEMS, Queensland University of Technology, Australia, n.ye@qut.edu.au 3Department of Computer Science, National University of Singapore, Singapore, leews@comp.nus.edu.sg
Pseudocode Yes Algorithm 1 Active learning for the mixture prior model
Open Source Code No The paper does not contain any explicit statements or links indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We run maximum Gibbs error with 1/σ2 = 0.01, 0.1, 0.2, 1, 10 on tasks from the 20 Newsgroups and UCI data sets (Joachims 1996; Bache and Lichman 2013) shown in the first column of Table 1.
Dataset Splits No The paper mentions 'randomly choose the first 10 examples as a seed set' and 'first 150 selected examples' for AUC calculation, and refers to a 'separate test set'. However, it does not specify any explicit training, validation, or test splits by percentage or absolute count, nor does it refer to predefined splits from cited works. It refers to 'a validation set' conceptually for passive learning, but not for its own experimental setup.
Hardware Specification No The paper does not provide any specific details regarding the hardware (e.g., CPU, GPU models, memory) used to conduct the experiments.
Software Dependencies No The paper mentions using a 'logistic regression model' but does not specify any software libraries, frameworks, or their version numbers that were used for implementation or experimentation.
Experiment Setup Yes We use the logistic regression model with different L2 regularizers... We run maximum Gibbs error with 1/σ2 = 0.01, 0.1, 0.2, 1, 10... For AL, we randomly choose the first 10 examples as a seed set. The scores are averaged over 100 runs of the experiments with different seed sets.