How Well Can Transformers Emulate In-context Newton's Method?
This work addresses the theoretical capabilities of Transformers for in-context learning, providing insights into their algorithmic potential, though it is incremental as it builds on prior studies of optimization emulation.
The paper tackles the problem of whether Transformers can emulate higher-order optimization methods like Newton's method beyond linear regression, showing that linear attention Transformers with ReLU layers can approximate second-order algorithms for logistic regression with ε error using logarithmic layers and implement a step of Newton's iteration for matrix inversion with two layers.
Transformer-based models have demonstrated remarkable in-context learning capabilities, prompting extensive research into its underlying mechanisms. Recent studies have suggested that Transformers can implement first-order optimization algorithms for in-context learning and even second order ones for the case of linear regression. In this work, we study whether Transformers can perform higher order optimization methods, beyond the case of linear regression. We establish that linear attention Transformers with ReLU layers can approximate second order optimization algorithms for the task of logistic regression and achieve $ε$ error with only a logarithmic to the error more layers. As a by-product we demonstrate the ability of even linear attention-only Transformers in implementing a single step of Newton's iteration for matrix inversion with merely two layers. These results suggest the ability of the Transformer architecture to implement complex algorithms, beyond gradient descent.