Parallel Context-of-Experts Decoding for Retrieval Augmented Generation
This addresses efficiency and reasoning challenges in retrieval-augmented generation systems, though it appears incremental as it builds on existing methods without introducing a new paradigm.
The paper tackles the trade-off in Retrieval Augmented Generation between multi-document reasoning and computational bottlenecks by proposing Parallel Context-of-Experts Decoding, a training-free framework that shifts evidence aggregation to decoding, enabling cross-document interaction without shared attention.
Retrieval Augmented Generation faces a trade-off: concatenating documents in a long prompt enables multi-document reasoning but creates prefill bottlenecks, while encoding document KV caches separately offers speed but breaks cross-document interaction. We propose Parallel Context-of-Experts Decoding (Pced), a training-free framework that shifts evidence aggregation from the attention mechanism to the decoding. Pced treats retrieved documents as isolated "experts", synchronizing their predictions via a novel retrieval-aware contrastive decoding rule that weighs expert logits against the model prior. This approach recovers cross-document reasoning capabilities without constructing a shared attention across documents.