Imbalanced Label Distribution Learning
Authors: Xingyu Zhao, Yuexuan An, Ning Xu, Jing Wang, Xin Geng
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments verify the superior performance of RDA. |
| Researcher Affiliation | Academia | 1School of Computer Science and Engineering, Southeast University, Nanjing 211189, China 2Key Laboratory of Computer Network and Information Integration (Ministry of Education), Southeast University, Nanjing 211189, China |
| Pseudocode | No | The paper describes methods and algorithms but does not contain a clearly labeled pseudocode or algorithm block. |
| Open Source Code | Yes | Details of the further analyses are provided in the appendix, which is available at: https://github.com/ailearn-ml/RDA. |
| Open Datasets | Yes | We curate six ILDL benchmarks that span movie rating, facial beauty perception and visual sentiment distribution perception. These datasets are sampled from six standard LDL datasets, including Movie (Geng 2016), SCUT-FBP (Xie et al. 2015), Emotion6 (Peng et al. 2015), Flickr LDL (Yang, Sun, and Sun 2017), RAF-ML (Shang and Deng 2019) and Natural Scene (Geng 2016). |
| Dataset Splits | Yes | The sampling process is performed 10 times, for each time, we sample the training set and validation set from the original training set which occupies 90% of the examples, while the test set remains unchanged. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU models, CPU types, memory) used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9). |
| Experiment Setup | Yes | In RDA, gϵ, hφ and Fϑ are set as linear projections, Gε and Hϕ are set as single-layer neural network with two outputs including mean and variance of Gaussian, and the modified distribution-balanced focal loss is adopted as the loss function V . Hyperparameters λ1, λ2 and λ3 are selected by grid search from the set {0.01, 0.05, 0.1, 0.2, 0.5}. |