Modelling Class Noise with Symmetric and Asymmetric Distributions
Authors: Jun Du, Zhihua Cai
AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The empirical study shows that, the proposed asymmetric models overall outperform the benchmark linear models, and the asymmetric Laplace-noise model achieves the best performance among all. |
| Researcher Affiliation | Academia | Jun Du School of Computer Science China University of Geosciences Wuhan, P. R. China, 430074 dr.jundu@gmail.com Zhihua Cai School of Computer Science China University of Geosciences Wuhan, P. R. China, 430074 zhcai@cug.edu.cn |
| Pseudocode | Yes | Algorithm 1 Learning asymmetric models |
| Open Source Code | No | The paper states: 'The raw annotation data is provided at https : //sites.google.com/site/nlpannotations/.' This link is for data used in assumption verification, not for the authors' source code. |
| Open Datasets | Yes | Empirical study is conducted on synthetic data and real-world UCI (Bache and Lichman 2013) data. |
| Dataset Splits | Yes | Grid-search on regularization coefficients using 10-fold cross-validation is applied (Hsu, Chang, and Lin 2010). |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., CPU, GPU models, memory, or cloud instances) used for running the experiments. |
| Software Dependencies | No | The paper references libraries like 'LIBLINEAR' and 'scikit-learn package' and mentions Python, but does not provide specific version numbers for any software dependencies. |
| Experiment Setup | Yes | More specifically, we assume a zero mean isotropic Gaussian prior on w: p(w) = N(w|0, α^-1I), where α is the precision parameter, and I is the identity matrix. We also assume a Gamma prior (with shape parameter α and scale parameter β ) on λ: p(λ) = Gamma(α , β ). We further set the mode of the Gamma prior to 1, such that the symmetric class-noise models (where λ = 1) are preferred: α-1/β = 1 => α = β + 1. The prior on λ therefore can be formulated: p(λ) = Gamma(β + 1, β) where β > 0. For all the asymmetric models, we initialize λ to 1, and w to the final solutions of the symmetric counterparts. |