Neural Collapse Inspired Feature-Classifier Alignment for Few-Shot Class-Incremental Learning

Authors: Yibo Yang, Haobo Yuan, Xiangtai Li, Zhouchen Lin, Philip Torr, Dacheng Tao

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on the mini Image Net, CUB-200, and CIFAR-100 datasets demonstrate that our proposed framework outperforms the state-of-the-art performances. Code address: https://github.com/Neural Collapse Applications/FSCIL
Researcher Affiliation Collaboration 1JD Explore Academy 2School of Computer Science, Wuhan University 3National Key Lab of General AI, School of Intelligence Science and Technology, Peking University 4Institute for Artificial Intelligence, Peking University 5Peng Cheng Laboratory 6University of Oxford
Pseudocode No The paper does not contain an explicit pseudocode block or a section formally labeled as an algorithm.
Open Source Code Yes Code address: https://github.com/Neural Collapse Applications/FSCIL
Open Datasets Yes We conduct our experiments on three FSCIL benchmark datasets including mini Image Net (Russakovsky et al., 2015), CIFAR-100 (Krizhevsky et al., 2009), and CUB-200 (Wah et al., 2011).
Dataset Splits No The paper describes the data distribution for sessions (e.g., 5-way 5-shot) and the classes used for training and evaluation. However, it does not explicitly mention a separate "validation set" or "validation split" for model selection or hyperparameter tuning during training, which is a key requirement for this question.
Hardware Specification No The paper mentions the use of ResNet architectures (e.g., Res Net-12, Res Net-18) as backbones but does not specify any hardware details such as GPU models, CPU types, or memory used for running the experiments.
Software Dependencies No The paper mentions using SGD with momentum as an optimizer and a cosine annealing strategy for learning rate. However, it does not provide specific version numbers for any software dependencies or libraries (e.g., Python 3.x, PyTorch 1.x).
Experiment Setup Yes We train all models with a batchsize of 512 in the base session, and a batchsize of 64 (containing new session data and intermediate features in the memory) in each incremental session. On mini Image Net, we train for 500 epochs in the base session, and 100-170 iterations in each incremental session. The initial learning rate is 0.25 for base session, and 0.025 for incremental sessions.