Summary of Mitigating Catastrophic Forgetting in Large Language Models with Self-synthesized Rehearsal, by Jianheng Huang et al.
Mitigating Catastrophic Forgetting in Large Language Models with Self-Synthesized Rehearsal
by Jianheng Huang, Leyang Cui, Ante Wang, Chengyi Yang, Xinting Liao, Linfeng Song, Junfeng Yao, Jinsong Su
First submitted to arxiv on: 2 Mar 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
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 A novel framework called Self-Synthesized Rehearsal (SSR) is proposed for continual learning of large language models (LLMs). Conventional rehearsal-based methods rely on access to original training data, which may not be feasible in real-world applications. SSR leverages the base LLM to generate synthetic instances for rehearsal, and then refines these instances using the latest LLM. This approach achieves superior or comparable performance to conventional methods while being more data-efficient. Experimental results demonstrate that SSR effectively preserves the generalization capabilities of LLMs in general domains. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models can forget things they learned earlier if they don’t get practiced enough. Scientists have found a way to make these models remember what they learned by using their own “thoughts” as practice exercises. This new approach is called Self-Synthesized Rehearsal, or SSR for short. Instead of needing all the original training data, SSR lets the model generate its own practice problems and then learn from those. In tests, this method worked just as well or better than other methods that need more data. It’s like a way to help the model “keep in shape” without needing everything it learned before. |
Keywords
» Artificial intelligence » Continual learning » Generalization
