Summary of Low-dimension-to-high-dimension Generalization and Its Implications For Length Generalization, by Yang Chen et al.
Low-Dimension-to-High-Dimension Generalization And Its Implications for Length Generalization
by Yang Chen, Yitao Liang, Zhouchen Lin
First submitted to arxiv on: 11 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 paper investigates Low-Dimension-to-High-Dimension (LDHD) generalization, a specific type of Out-of-Distribution (OOD) generalization. The authors demonstrate theoretically that LDHD generalization is generally unattainable without incorporating prior knowledge to provide inductive bias. They explore LDHD generalization in Boolean functions and show that different architectures trained with Stochastic Gradient Descent (SGD) converge to min-degree interpolators w.r.t. different independent sets. The authors also propose a principle for position embedding design to handle both LDHD generalization and nuisances like data format, leading to the development of a novel position embedding called RPE-Square. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about a special kind of learning problem where we try to apply what we learned in a simple space to a more complex one. The authors show that this can be very challenging without some extra help. They look at how different approaches work in Boolean functions, which are like true or false statements. They also come up with new ideas for making it easier to learn from different kinds of data. |
Keywords
» Artificial intelligence » Embedding » Generalization » Stochastic gradient descent
