SCALE: Small-Model-First Concept Abstraction with LLM-Guided Experience Internalization
SCALE: Small-Model-First Concept Abstraction with LLM-Guided Experience Internalization
This research paper proposes SCALE, a small-model-first framework that integrates selective large language model (LLM) reasoning, experience-driven concept abstraction, and knowledge internalization. The system allows lightweight models to handle routine predictions while using the LLM for reflection on prediction errors and accumulated evidence. Validated concepts are incorporated into hierarchical neural representations, improving held-out accuracy from 87.11% to 98.28% and difficult-case accuracy from 55.33% to 66.47% across ten random seeds, while reducing active experience records from 6,000 to 1,528.
The paper introduces SCALE, a framework that uses selective LLM reasoning to abstract experience into compact knowledge for lightweight models, reporting specific accuracy improvements on held-out data across ten random seeds.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Control
- Reported autonomousOpen source S2
Reported numbers
Reported accuracy
66.47%
View original evidence
cy from 87.11% to 98.28%, hierarchical concept knowledge increases difficult-case accuracy from 55.33% to 66.47%, and abstract archiving reduces active experience records from 6,000 to 1,528.
Open source S5Reported accuracy
55.33%
View original evidence
cy from 87.11% to 98.28%, hierarchical concept knowledge increases difficult-case accuracy from 55.33% to 66.47%, and abstract archiving reduces active experience records from 6,000 to 1,528.
Open source S5random seeds count
10
View original evidence
concept hypotheses. Independently validated concepts are progressively incorporated into hierarchical neural representations, enabling lightweight models to exploit increasingly abstract knowledge. Meanwhile, mastered experiences are excluded from active memory while their original records remain available in a complete archive. Experiments across ten random seeds show that historical recall improves held-out accura
Open source S4
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
arXiv:2611.12120v1 Announce Type: cross Abstract: Autonomous agents must continually acquire reusable knowledge from experience without relying on costly large language model (LLM) reasoning for every decision. However, existing experience-based approaches often struggle to transform accumulated observations and reasoning into compact knowledge that lightweight models can independently apply. This paper proposes SCAL
Open source S2
cy from 87.11% to 98.28%, hierarchical concept knowledge increases difficult-case accuracy from 55.33% to 66.47%, and abstract archiving reduces active experience records from 6,000 to 1,528.
Open source S5
concept hypotheses. Independently validated concepts are progressively incorporated into hierarchical neural representations, enabling lightweight models to exploit increasingly abstract knowledge. Meanwhile, mastered experiences are excluded from active memory while their original records remain available in a complete archive. Experiments across ten random seeds show that historical recall improves held-out accura
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org