Summary of Contrastive Augmented Graph2graph Memory Interaction For Few Shot Continual Learning, by Biqing Qi et al.
Contrastive Augmented Graph2Graph Memory Interaction for Few Shot Continual Learning
by Biqing Qi, Junqi Gao, Xingquan Chen, Dong Li, Jianxing Liu, Ligang Wu, Bowen Zhou
First submitted to arxiv on: 7 Mar 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary Few-Shot Class-Incremental Learning (FSCIL) is a crucial problem in machine learning that addresses continuously arriving classes. However, this task encounters additional challenges due to the scarcity of samples in new sessions, which intensifies overfitting and catastrophic forgetting. To mitigate these issues, existing methods rely on Explicit Memory (EM), comprising class prototypes. These methods perform Vector-to-Vector (V2V) interaction between features and prototypes stored in EM, neglecting local feature relationships. This paper proposes an extension to V2V interaction called Graph-to-Graph (G2G) interaction to incorporate local geometric structure information. Additionally, the Local Graph Preservation (LGP) mechanism is introduced to prevent local feature collapse and promote G2G alignment. To address sample scarcity in new classes, Contrast-Augmented G2G (CAG2G) promotes same-class feature aggregation for few-shot learning. The proposed method outperforms existing methods on CIFAR100, CUB200, and ImageNet-R datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about a way to improve learning when new classes arrive frequently. This is important because it helps machines learn from new information without forgetting old things. The problem is that there are too few examples of the new classes, which makes it hard to avoid mistakes and remember old information. The solution involves using geometric structures in local features to help with this problem. The method also includes ways to prevent the loss of important details and promote learning from new information. The results show that this approach works better than existing methods on different datasets. |
Keywords
» Artificial intelligence » Alignment » Few shot » Machine learning » Overfitting
