Generative Label Enhancement with Gaussian Mixture and Partial Ranking
Authors: Yunan Lu, Liang He, Fan Min, Weiwei Li, Xiuyi Jia
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, extensive experiments on real-world datasets validate our proposal. |
| Researcher Affiliation | Academia | 1 School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, China 2 School of Computer Science, Southwest Petroleum University, Chengdu, China 3 College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing, China 4 Ministry Key Laboratory for Safety-Critical Software Development and Verification, Nanjing University of Aeronautics and Astronautics, Nanjing, China |
| Pseudocode | No | The paper describes the generative and inference models using equations and graphical representations (Figure 2), but it does not include any pseudocode or explicitly labeled algorithm blocks. |
| Open Source Code | No | The paper does not contain any statement about making the source code available, nor does it provide a link to a code repository. |
| Open Datasets | Yes | The datasets we used is shown in Table 1. These datasets come from the tasks including emotion mining (No. 1-5), natural scene predicting (No. 6), movie rating predicting (No. 7), and bioinformatics (No. 8-14)1. 1Although there are 10 Yeast datasets released by Geng (2016), we only select those with more than 4 labels due to page limitation. twit , flic (Yang, Sun, and Sun 2017) and emo6 (Peng et al. 2015) datasets |
| Dataset Splits | No | Then, we randomly partition dataset (90% for training and 10% for testing). The paper specifies a 90% training and 10% testing split, but no explicit validation split or cross-validation setup is mentioned. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU models, CPU types, memory) used to run the experiments. |
| Software Dependencies | No | The paper mentions using 'SABFGS (Geng 2016)' and the 'SGVB (Kingma and Welling 2014) framework', but it does not provide specific version numbers for any software dependencies, libraries, or programming languages. |
| Experiment Setup | Yes | For our method, we set K = 3, λy = 105, and λz is selected in {10 3, 10 2, 10 1, 100}; both µx( ) and µd( ) are modeled as linear functions; both σ2 x( ) and σ2 d( ) are modeled as linear functions with the softplus activation function. Besides, to accelerate the convergence, we use min-max normalization to preprocess the feature matrices of all datasets. As suggested by Kingma and Welling (2014), we generate only one Monte Carlo sample for each observation, i.e., S = 1. For LEVI, the MLPs are constructed with two hidden layers, each with 500 hidden units and softplus activation function, i.e., ln(1+exp( )). |