Summary of Distributional Associations Vs In-context Reasoning: a Study Of Feed-forward and Attention Layers, by Lei Chen et al.
Distributional Associations vs In-Context Reasoning: A Study of Feed-forward and Attention Layers
by Lei Chen, Joan Bruna, Alberto Bietti
First submitted to arxiv on: 5 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (stat.ML)
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 differences between feed-forward and attention layers in large language models, particularly in terms of knowledge storage and in-context reasoning. The authors analyze a controlled synthetic setting where certain predictions involve both distributional and contextual information, finding that feed-forward layers tend to learn simple associations while attention layers focus on context-dependent reasoning. Theoretical analysis reveals the role of gradient noise in this discrepancy. The study also ablates pre-trained models on simple reasoning tasks, demonstrating similar disparities. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how language models process information and make predictions. It compares two types of layers: feed-forward and attention. These layers are important for making connections between words and understanding the context of a sentence. Researchers created a special test to see which layer is better at learning simple patterns or more complex ideas. They found that one type of layer is good at finding simple relationships, while the other is better at figuring out what’s going on in a specific situation. |
Keywords
» Artificial intelligence » Attention
