Transformer Memory as a Differentiable Search Index
This presents a new paradigm for information retrieval that could simplify systems for users and developers, though it is incremental in applying Transformers to retrieval tasks.
The paper tackles information retrieval by introducing the Differentiable Search Index (DSI), a Transformer-based model that encodes corpus information in its parameters to map queries directly to document IDs, simplifying retrieval and outperforming strong baselines like dual encoder models and BM25 in zero-shot setups.
In this paper, we demonstrate that information retrieval can be accomplished with a single Transformer, in which all information about the corpus is encoded in the parameters of the model. To this end, we introduce the Differentiable Search Index (DSI), a new paradigm that learns a text-to-text model that maps string queries directly to relevant docids; in other words, a DSI model answers queries directly using only its parameters, dramatically simplifying the whole retrieval process. We study variations in how documents and their identifiers are represented, variations in training procedures, and the interplay between models and corpus sizes. Experiments demonstrate that given appropriate design choices, DSI significantly outperforms strong baselines such as dual encoder models. Moreover, DSI demonstrates strong generalization capabilities, outperforming a BM25 baseline in a zero-shot setup.