A Control Architecture for Training-Free Memory Use

TL;DR

This paper introduces a control architecture for training-free memory use in language models, improving reasoning accuracy by selectively applying retrieved memory content based on confidence and uncertainty measures.

Contribution

It presents a novel applicability control method combining uncertainty routing, confidence-based acceptance, and memory bank management, enhancing reasoning benchmarks without model training.

Findings

Improves SVAMP by +7.0 points and ASDiv by +7.67 points over baseline.

Control architecture, not raw memory exposure, drives improvements.

Confidence measures effectively distinguish helpful from harmful memory interventions.

Abstract

Prompt-injected memory can improve reasoning without updating model weights, but it also creates a control problem: retrieved content helps only when it is applied in the right state. We study this problem in a strict training-free setting and formulate it as applicability control: when to trigger a memory-assisted second pass, when to trust it, and how to maintain the memory bank over time. Our method combines uncertainty-based routing, confidence-based selective acceptance, bank selection across rule and exemplar memory, and evidence-based governance of the memory bank over time. Under a locked training-free protocol with compute-matched controls, it improves two core arithmetic benchmarks by +7.0 points on SVAMP and +7.67 points on ASDiv over baseline. The same architecture also transfers to QA and agent benchmarks with smaller positive effects and shows the same positive direction on a second checkpoint for the main arithmetic tasks. On arithmetic, the main empirical pattern is that the control architecture, rather than raw memory exposure, drives the improvements on SVAMP and ASDiv. Mechanistically, confidence separates helpful from harmful rule-bank interventions, and under fixed retrieval the repair-versus-corrupt difference localizes to rows whose retrieved set actually contains the edited entries.