Polynomial Regression as a Task for Understanding In-context Learning Through Finetuning and Alignment
This work addresses a foundational problem for researchers studying in-context learning in large language models, but it is incremental as it builds on existing toy problem approaches.
The authors tackled the problem of understanding in-context learning in transformers by proposing univariate polynomial regression as a toy function class, which allows for clearer visualization and study of prompting and alignment compared to simpler classes like linear regression.
Simple function classes have emerged as toy problems to better understand in-context-learning in transformer-based architectures used for large language models. But previously proposed simple function classes like linear regression or multi-layer-perceptrons lack the structure required to explore things like prompting and alignment within models capable of in-context-learning. We propose univariate polynomial regression as a function class that is just rich enough to study prompting and alignment, while allowing us to visualize and understand what is going on clearly.