On the Optimal Memorization Capacity of Transformers
This provides foundational insights into the efficiency of Transformers for memorization tasks, relevant for researchers in machine learning and neural network theory.
The paper tackles the problem of understanding the memorization capacity of Transformers, showing they can memorize labels with $ ilde{O}(\sqrt{N})$ parameters in next-token prediction, which is optimal up to logarithmic factors, and $ ilde{O}(\sqrt{nN})$ parameters are necessary and sufficient in sequence-to-sequence settings.
Recent research in the field of machine learning has increasingly focused on the memorization capacity of Transformers, but how efficient they are is not yet well understood. We demonstrate that Transformers can memorize labels with $\tilde{O}(\sqrt{N})$ parameters in a next-token prediction setting for $N$ input sequences of length $n$, which is proved to be optimal up to logarithmic factors. This indicates that Transformers can efficiently perform memorization with little influence from the input length $n$ owing to the benefit of parameter sharing. We also analyze the memorization capacity in the sequence-to-sequence setting, and find that $\tilde{O}(\sqrt{nN})$ parameters are not only sufficient, but also necessary at least for Transformers with hardmax. These results suggest that while self-attention mechanisms can efficiently identify input sequences, the feed-forward network becomes a bottleneck when associating a label to each token.