BEEF: Bi-Compatible Class-Incremental Learning via Energy-Based Expansion and Fusion
Authors: Fu-Yun Wang, Da-Wei Zhou, Liu Liu, Han-Jia Ye, Yatao Bian, De-Chuan Zhan, Peilin Zhao
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on three widely used benchmarks: CIFAR-100, Image Net-100, and Image Net-1000 demonstrate that BEEF achieves state-of-the-art performance in both the ordinary and challenging CIL settings. |
| Researcher Affiliation | Collaboration | 1State Key Laboratory for Novel Software Technology, Nanjing University 2Tencent AI Lab |
| Pseudocode | No | The paper describes the training framework, model expansion, and fusion process in detail across several sections and an appendix. However, it does not include any clearly labeled pseudocode blocks or algorithms. |
| Open Source Code | Yes | The Code is available at https://github.com/G-U-N/ICLR23-BEEF. |
| Open Datasets | Yes | We validate our methods on widely used benchmarks of class-incremental learning CIFAR100 (Krizhevsky et al., 2009) and Image Net100/1000 (Deng et al., 2009). CIFAR-100: CIFAR-100 consists of 50,000 training images... Image Net-1000: Image Net-1000 is a large scale dataset composed of about 1.28 million images for training... |
| Dataset Splits | Yes | Image Net-1000: Image Net-1000 is a large scale dataset composed of about 1.28 million images for training and 50,000 for validation with 500 images per class. |
| Hardware Specification | No | The paper states that experiments were implemented with PyTorch and ran on certain datasets, and provides training time and peak memory usage in tables. However, it does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments. |
| Software Dependencies | No | Our method and all baselines are implemented with Pytorch (Paszke et al., 2017) in Py CIL (Zhou et al., 2021a). While software names are mentioned, specific version numbers for PyTorch or Py CIL are not provided, which prevents full reproducibility of the software environment. |
| Experiment Setup | Yes | For Image Net, we adopt the standard Res Net-18... and set the batch size as 256. The learning rate starts from 0.1 and gradually decays at milestones (170 epochs in total)... For both Image Net and CIFAR-100, we use SGD with the momentum of 0.9 and the weight decay of 5e-4 at the expansion phase. At the fusion phase, we use SGD with a momentum of 0.9 and set the weight decay to 0, and train the fused model on the exemplar-set for 60 epochs. |