Superiority of Multi-Head Attention in In-Context Linear Regression
This provides theoretical justification for the multi-head attention design in transformers, which is incremental but addresses a known bottleneck in understanding attention mechanisms for machine learning practitioners.
The paper theoretically analyzes multi-head versus single-head attention in transformers for in-context linear regression, showing that multi-head attention achieves better performance with a smaller multiplicative constant in prediction loss as the number of examples increases.
We present a theoretical analysis of the performance of transformer with softmax attention in in-context learning with linear regression tasks. While the existing literature predominantly focuses on the convergence of transformers with single-/multi-head attention, our research centers on comparing their performance. We conduct an exact theoretical analysis to demonstrate that multi-head attention with a substantial embedding dimension performs better than single-head attention. When the number of in-context examples D increases, the prediction loss using single-/multi-head attention is in O(1/D), and the one for multi-head attention has a smaller multiplicative constant. In addition to the simplest data distribution setting, we consider more scenarios, e.g., noisy labels, local examples, correlated features, and prior knowledge. We observe that, in general, multi-head attention is preferred over single-head attention. Our results verify the effectiveness of the design of multi-head attention in the transformer architecture.