Summary of Hybrid Memory Replay: Blending Real and Distilled Data For Class Incremental Learning, by Jiangtao Kong et al.
Hybrid Memory Replay: Blending Real and Distilled Data for Class Incremental Learning
by Jiangtao Kong, Jiacheng Shi, Ashley Gao, Shaohan Hu, Tianyi Zhou, Huajie Shao
First submitted to arxiv on: 20 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- 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 The proposed hybrid memory optimization combines the strengths of synthetic and real exemplars to mitigate catastrophic forgetting in Class Incremental Learning (CIL) when the buffer size for exemplars is limited. The innovative modification to Data Distillation (DD) distills synthetic data from a sliding window of checkpoints in history, which is then conditioned on to select complementary real exemplars. This approach effectively outperforms existing replay-based baselines across multiple benchmarks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes a new way to store memories when learning new tasks while keeping the knowledge learned from previous tasks. It combines two ideas: one that stores synthetic data and another that stores real data. The goal is to help the model remember what it learned before, even if there’s only limited space for storing old information. This approach can be used with most existing models that learn in this way. |
Keywords
* Artificial intelligence * Distillation * Optimization * Synthetic data
