Amazon's BoundaryMORPH uses Gaussian Processes to budget cross-encoder reranking, +5.4 nCG@100
_reachsumit · x · 2026-09-24
Amazon proposes BoundaryMORPH, an algorithm for allocating cross-encoder (CE) reranking budgets in RAG with diffuse, open-ended queries.
- Problem: the CE budget B is strictly latency-bound and often smaller than the LLM context capacity k. Standard reranking wastes compute verifying obvious top candidates while ignoring relevant documents further down the initial ranking.
- Method: using a Gaussian Process, BoundaryMORPH treats the dual-encoder ranking as a structural prior and spends CE calls on resolving top-k set membership at the boundary rather than finding a single most-relevant document; information from each CE call propagates to unscored documents, maximizing budget utility.
- Results: state-of-the-art set retrieval quality across multiple models and datasets, +5.4 nCG@100 over the strongest baseline.
More from Research
- Anthropic's enzyme research uses AlphaFold, showing LLMs and specialized models complement each other — JMateosGarcia · 2026-09-24
- Claude teams up with AlphaFold for novel enzyme research, showing LLM-specialized model synergy — JMateosGarcia · 2026-09-24
- ECCV 2026 paper: x0-prediction fixes inefficient diffusion in reconstruction-tuned RAE latent spaces — serrjoa · 2026-09-24
- RecCAR closes reciprocal cross-attention gap in joint video diffusion models — barilan · 2026-09-24
- Berkeley's Do as I Do turns everyday human videos into dexterous robot hand training data — micoolcho · 2026-09-24
- Weaving directly shaped math's conceptual vocabulary, argues cognitive science thread — abenitezburraco · 2026-09-24