Apple researchers propose learned networks to speed up RAG retrieval
Overview
Apple researchers present SupportNet and KeyNet, small neural networks trained offline to route queries to their best-matching keys faster than centroid-based search, according to a paper at ICML 2026 titled "Amortizing Maximum Inner Product Search with Learned Support Functions." The paper says the approach trades an upfront training cost for faster, more accurate retrieval at inference time. The networks are trained once for a fixed database of keys. The report does not give benchmark figures for the speedup or accuracy gain.
Written by AI from the articles below · updated Nov 5, 1:09 PM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- Apple Machine Learning ResearchOfficialApple researchers train learned networks to speed up RAG retrieval
AIApple researchers present SupportNet and KeyNet, small neural networks trained offline to route queries to their best-matching keys faster than centroid-based search, at ICML 2026. The paper, Amortizing Maximum Inner Product Search with Learned Support Functions, says the approach trades an upfront training cost for faster, more accurate retrieval at inference time.
Heat Trend
Not enough continuous observations to show a trend yet.