Looped Transformers for Length Generalization
This addresses a key limitation in Transformer models for iterative tasks, though it is incremental as it builds on known iterative solutions.
The paper tackled the problem of Transformers struggling with length generalization on arithmetic and algorithmic tasks by introducing looped Transformers with adaptive steps, resulting in significant improvements in handling inputs of unseen lengths.
Recent work has shown that Transformers trained from scratch can successfully solve various arithmetic and algorithmic tasks, such as adding numbers and computing parity. While these Transformers generalize well on unseen inputs of the same length, they struggle with length generalization, i.e., handling inputs of unseen lengths. In this work, we demonstrate that looped Transformers with an adaptive number of steps significantly improve length generalization. We focus on tasks with a known iterative solution, involving multiple iterations of a RASP-L operation - a length-generalizable operation that can be expressed by a finite-sized Transformer. We train looped Transformers using our proposed learning algorithm and observe that they learn highly length-generalizable solutions for various tasks.