Bayesian Max-margin Multi-Task Learning with Data Augmentation

Authors: Chengtao Li, Jun Zhu, Jianfei Chen

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical results demonstrate superior performance than competitors in both multi-task classification and regression. and 6. Experiments We present empirical studies for both multi-task classification and multi-task regression.
Researcher Affiliation Academia Chengtao Li CTLI.CS@HOTMAIL.COM Jun Zhu DCSZJ@MAIL.TSINGHUA.EDU.CN Jianfei Chen CHENJF10@MAILS.TSINGHUA.EDU.CN Institute for Interdisciplinary Information Sciences, Tsinghua University, Beijing, China Dept. of Comp. Sci. & Tech, TNList Lab, State Key Lab of Intell. Tech & Sys, Tsinghua University, Beijing, China
Pseudocode No The paper describes the Gibbs sampling algorithms through textual explanation and mathematical equations but does not include structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code for the methodology described, nor does it mention code availability via a repository link or supplementary materials.
Open Datasets Yes We use four multi-label datasets that are publicly available1. Table 1 summarizes their statistics. For multi-label classification, we formulate it as a multi-task learning problem, where each task is a binary classifier determining whether an instance has a particular label. 1http://mulan.sourceforge.net/datasets. html and We use the public School dataset, which consists of the examination records of 15,362 students from 139 secondary schools in years 1985, 1986 and 1987. The dataset has been used to study the effectiveness of schools. It has been used extensively to evaluate multi-task learning methods (Bakker & Heskes, 2003; Zhang & Yeung, 2012).
Dataset Splits Yes The regularization parameter C is chosen from [10 3, 103] using 5-fold cross validation.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details, such as library or solver names with version numbers, needed to replicate the experiment.
Experiment Setup Yes The regularization parameter C is chosen from [10 3, 103] using 5-fold cross validation. and We empirically set ε to be 0.001. The hyper-parameter of the IBP prior of Z (i.e., α) is fixed at 5 in this experiment, and we will return to investigate the sensitivity over this parameter in Section 6.3.2. We search the regularization parameter C in the range of [0.1,10] with 5-fold cross validation.