Optimality and NP-Hardness of Transformers in Learning Markovian Dynamical Functions
This work addresses a theoretical gap in transformer capabilities for dynamics-driven tasks, though it is incremental as it builds on prior linear regression studies.
The paper tackles the problem of understanding how transformers perform in-context learning for Markovian dynamical functions, revealing that finding optimal parameters for a single-layer linear self-attention model is NP-hard, which limits its ability to represent structured dynamics.
Transformer architectures can solve unseen tasks based on input-output pairs in a given prompt due to in-context learning (ICL). Existing theoretical studies on ICL have mainly focused on linear regression tasks, often with i.i.d. inputs. To understand how transformers express ICL when modeling dynamics-driven functions, we investigate Markovian function learning through a structured ICL setup, where we characterize the loss landscape to reveal underlying optimization behaviors. Specifically, we (1) provide the closed-form expression of the global minimizer (in an enlarged parameter space) for a single-layer linear self-attention (LSA) model; (2) prove that recovering transformer parameters that realize the optimal solution is NP-hard in general, revealing a fundamental limitation of one-layer LSA in representing structured dynamical functions; and (3) supply a novel interpretation of a multilayer LSA as performing preconditioned gradient descent to optimize multiple objectives beyond the square loss. These theoretical results are numerically validated using simplified transformers.