Joint Input and Output Coordination for Class-Incremental Learning
Authors: Shuai Wang, Yibing Zhan, Yong Luo, Han Hu, Wei Yu, Yonggang Wen, Dacheng Tao
IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments show that our mechanism can significantly improve their performance. To demonstrate the effectiveness of our mechanism, we incorporate it into some recent or competitive incremental learning approaches on multiple popular datasets (CIFAR10-LT, CIFAR100-LT, CIFAR100 [Krizhevsky et al., 2009], Mini Imag Net [Vinyals et al., 2016], Tiny Image Net [Le and Yang, 2015] and Cub200-2011 [Wah et al., 2011]). The results show that we can consistently improve the existing approaches, and the relative improvement is more than 10% sometimes. |
| Researcher Affiliation | Collaboration | Shuai Wang1,2 , Yibing Zhan3 , Yong Luo1,2 , Han Hu4 , Wei Yu1 , Yonggang Wen5 , Dacheng Tao5 1Institute of Artificial Intelligence, School of Computer Science, Wuhan University, China 2 Hubei Luojia Laboratory, Wuhan, China 3JD Explore Academy, JD.com, Inc., China 4School of Information and Electronics, Beijing Institute of Technology, China 5College of Computing & Data Science, Nanyang Technological University, Singapore |
| Pseudocode | Yes | Algorithm 1 Main procedure of input coordination. Input: The data of the incremental learning model {Dt old, Dt}; the feature extractor of the current model is f ( , Θ); the parameter of the current fully-connected layer is W; Output: The updated parameters Θ and W; |
| Open Source Code | No | The paper does not provide any explicit statements about making its source code available (e.g., a link to a repository or a statement about supplementary material containing code). |
| Open Datasets | Yes | In this paper, we not only validate the effectiveness of our method on unbalanced CIFAR10-LT and CIFAR100-LT datasets but also conduct corresponding validation on balanced CIFAR100 [Krizhevsky et al., 2009], Mini Image Net [Vinyals et al., 2016], Tiny Image Net [Le and Yang, 2015], and Cub-200-2011 [Wah et al., 2011] datasets. |
| Dataset Splits | No | The paper mentions '50 validation images' for Tiny Image Net, and 'The training and test split of this dataset is 80 : 20' for Mini Image Net. For CIFAR datasets, it mentions '50, 000 training images and 10, 000 test images'. However, it does not consistently specify clear training/test/validation splits (e.g., percentages or exact counts for all splits) across all datasets to ensure full reproducibility of data partitioning, especially for validation. |
| Hardware Specification | Yes | We run the training on two NVIDIA 3090RTX GPUs. |
| Software Dependencies | No | Our method and all the compared approaches...are implemented using Py CIL [Zhou et al., 2023] and Pytorch [Paszke ets al., 2017]. The paper mentions software names but does not provide specific version numbers for Py CIL or Pytorch to ensure reproducibility. |
| Experiment Setup | Yes | In terms of parameter settings, we align with the original methods on Py CIL [Zhou et al., 2023] to facilitate a fair comparison. Among these, the batch size is set to 128. Additionally, the SGD optimizer is used to gradually update the weights during incremental learning model training. The learning rate is initially set to be 0.1 and gradually decays. |