Summary of Intelligent Learning Rate Distribution to Reduce Catastrophic Forgetting in Transformers, by Philip Kenneweg et al.
Intelligent Learning Rate Distribution to reduce Catastrophic Forgetting in Transformers
by Philip Kenneweg, Alexander Schulz, Sarah Schröder, Barbara Hammer
First submitted to arxiv on: 27 Mar 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 paper investigates the issue of catastrophic forgetting in transformer neural networks, which arises when language models are fine-tuned on different tasks. The authors challenge the conventional approach of using a flat learning rate for the entire network and instead perform hyperparameter optimization to find better learning rate distributions. These optimized rates are combined and shown to improve performance on NLP benchmarks from the GLUE dataset, demonstrating generalizability across various tasks. |
Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how language models can forget what they learned when switching between different tasks. The usual way to fine-tune these models is with a single learning rate that works for everything. But what if this isn’t the best approach? The researchers try to find better ways to adjust the learning rate and show that their method does a better job of preventing forgetting. They tested it on many natural language processing tasks and found it worked well. |
Keywords
» Artificial intelligence » Hyperparameter » Natural language processing » Nlp » Optimization » Transformer