$\infty$-former: Infinite Memory Transformer
This addresses the bottleneck of long-term memory in Transformers for applications requiring extensive context, such as document-based dialogue, though it is an incremental improvement over existing efficient transformers.
The paper tackles the problem of Transformers' limited memory capacity by proposing the ∞-former, which uses continuous-space attention to achieve unbounded long-term memory with fixed computation, resulting in effective information retention in tasks like language modeling and dialogue generation.
Transformers are unable to model long-term memories effectively, since the amount of computation they need to perform grows with the context length. While variations of efficient transformers have been proposed, they all have a finite memory capacity and are forced to drop old information. In this paper, we propose the $\infty$-former, which extends the vanilla transformer with an unbounded long-term memory. By making use of a continuous-space attention mechanism to attend over the long-term memory, the $\infty$-former's attention complexity becomes independent of the context length, trading off memory length with precision. In order to control where precision is more important, $\infty$-former maintains "sticky memories" being able to model arbitrarily long contexts while keeping the computation budget fixed. Experiments on a synthetic sorting task, language modeling, and document grounded dialogue generation demonstrate the $\infty$-former's ability to retain information from long sequences.