Generative Calibration of Inaccurate Annotation for Label Distribution Learning
Authors: Liang He, Yunan Lu, Weiwei Li, Xiuyi Jia
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments Experimental Configuration Datasets Based on the annotation methodology of datasets, we classify the LDL datasets into three distinct groups. The specific datasets we use are presented in Table 1. ... The results of the recovery experiment are shown in Table 2. ... The accompanying Figure 2 illustrates some of the experimental results obtained from our model. ... Ablation Study. |
| Researcher Affiliation | Academia | Liang He1, Yunan Lu1,2, Weiwei Li3, Xiuyi Jia1,2* 1School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, China 2Guangxi Key Lab of Multi-source Information Mining & Security, Guangxi Normal University, Guilin, China 3College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing, China |
| Pseudocode | No | The paper describes the generative model steps in numbered lists (1-4) under 'Generative Model', but these are descriptions of the model's components and generation process, not a formal pseudocode block or algorithm. |
| Open Source Code | No | The paper does not contain any statement about making its source code publicly available or a link to a code repository. |
| Open Datasets | Yes | Datasets Based on the annotation methodology of datasets, we classify the LDL datasets into three distinct groups. The specific datasets we use are presented in Table 1. Subjective annotation datasets: These datasets are sourced from subjective annotation tasks, including emotion mining (No.1-2) and movie rating prediction (No.3). ... Biological experiment datasets: This category of datasets comprises three Yeast datasets (No.4-7). ... Ranking dataset: This type of dataset includes a natural scene dataset (No.8). ... No Datasets Instance Features Labels 1 SJAFFE (sj) 213 243 6 2 Twitter-LDL (twit) 10045 168 8 3 Movie (mov) 7755 1869 5 4 Yeast-heat (heat) 2465 24 6 5 Yeast-cold (cold) 2465 24 4 6 Yeast-spo (spo) 2465 24 6 7 Yeast-spo5 (spo5) 2465 24 3 8 Nature-Scence (ns) 2000 294 9 |
| Dataset Splits | No | Firstly, we randomly partition the dataset into a training set (90%) and a testing set (10%). The paper does not mention a separate validation set. |
| Hardware Specification | No | The paper does not specify any hardware details such as CPU, GPU, or memory used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers. |
| Experiment Setup | Yes | For Duo-LDL, the learning rate set to 0.05 and the weight decay cost set to 0.5. For LDL-LRR, λ and β are selected from 10{ 6, 5,..., 2, 1} and 10{ 3, 2,...,1,2}, respectively. ... We set S = 1, which means generating only one Monte Carlo sample for each observation (Kingma and Welling 2014). |