Long-Tailed Partial Label Learning by Head Classifier and Tail Classifier Cooperation
Authors: Yuheng Jia, Xiaorui Peng, Ran Wang, Min-Ling Zhang
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The experiments on the benchmarks demonstrate the proposed approach improves the accuracy of the SOTA methods by a substantial margin. |
| Researcher Affiliation | Academia | 1School of Computer Science and Engineering, Southeast University 2Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education, China 3Shenzhen Key Laboratory of Advanced Machine Learning and Applications, School of Mathematical Science, Shenzhen University, Shenzhen, China 4Guangdong Key Laboratory of Intelligent Information Processing, Shenzhen University, Shenzhen, China 5Key Laboratory of Computer Network and Information Integration (Southeast University), Ministry of Education, China |
| Pseudocode | Yes | Algorithm 1: Training Process the Proposed Method |
| Open Source Code | Yes | Code and data are available at : https://github.com/pruirui/HTC-LTPLL. |
| Open Datasets | Yes | Following the previous works (Wang et al. 2022a; Hong et al. 2023), we evaluated our method on the long-tailed versions of CIFAR10 and CIFAR100 and a realworld LT-PLL dataset PASCAL VOC. [...] The real-world LT-PLL dataset PASCAL VOC is constructed from PASCAL VOC 2007 (Everingham et al. 2010) by RECORDS (Hong et al. 2023). |
| Dataset Splits | No | The paper mentions that 'the model with the best performance on the validation dataset is taken as the final model', indicating use of a validation set, but does not specify its size or how it was split from the main dataset. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running experiments, such as GPU or CPU models. |
| Software Dependencies | No | The paper mentions software components like 'SGD' and 'Res Net' but does not provide specific version numbers for any software dependencies. |
| Experiment Setup | Yes | We used the 18-layer Res Net as the backbone. For all methods, they were trained for 800 epochs using SGD as the optimizer with momentum of 0.9 and weight decay of 0.001. The initial learning rate was set to 0.01, and divided by 10 after 700 epochs. We set batch size to 256 for CIFAR10 and CIFAR100 and 64 for PASCAL VOC. [...] For our method, we fixed the mixup hyper-parameter ζ = 4, µ = 0.6 and linearly ramp up reliable sample ratio α from 0.2 to 0.6 in first 50 epochs, dynamic balancing coefficient β from 0 to 0.9 in first 50 epochs for all experiments. When imbalance ratio γ 200, the logit adjustment coefficient τ was set to 1.2, otherwise, τ = 2. |