Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MAPLE: A Sub-Agent Architecture for Memory, Learning, and Personalization in Agentic AI Systems

Deepak Babu Piskala

TL;DR

This work proposes MAPLE (Memory-Adaptive Personalized LEarning), a principled decomposition where Memory handles storage and retrieval infrastructure; Learning extracts intelligence from accumulated interactions asynchronously; and Personalization applies learned knowledge in real-time within finite context budgets.

Abstract

Large language model (LLM) agents have emerged as powerful tools for complex tasks, yet their ability to adapt to individual users remains fundamentally limited. We argue this limitation stems from a critical architectural conflation: current systems treat memory, learning, and personalization as a unified capability rather than three distinct mechanisms requiring different infrastructure, operating on different timescales, and benefiting from independent optimization. We propose MAPLE (Memory-Adaptive Personalized LEarning), a principled decomposition where Memory handles storage and retrieval infrastructure; Learning extracts intelligence from accumulated interactions asynchronously; and Personalization applies learned knowledge in real-time within finite context budgets. Each component operates as a dedicated sub-agent with specialized tooling and well-defined interfaces. Experimental evaluation on the MAPLE-Personas benchmark demonstrates that our decomposition achieves a 14.6% improvement in personalization score compared to a stateless baseline (p < 0.01, Cohen's d = 0.95) and increases trait incorporation rate from 45% to 75% -- enabling agents that genuinely learn and adapt.

MAPLE: A Sub-Agent Architecture for Memory, Learning, and Personalization in Agentic AI Systems

TL;DR

This work proposes MAPLE (Memory-Adaptive Personalized LEarning), a principled decomposition where Memory handles storage and retrieval infrastructure; Learning extracts intelligence from accumulated interactions asynchronously; and Personalization applies learned knowledge in real-time within finite context budgets.

Abstract

Large language model (LLM) agents have emerged as powerful tools for complex tasks, yet their ability to adapt to individual users remains fundamentally limited. We argue this limitation stems from a critical architectural conflation: current systems treat memory, learning, and personalization as a unified capability rather than three distinct mechanisms requiring different infrastructure, operating on different timescales, and benefiting from independent optimization. We propose MAPLE (Memory-Adaptive Personalized LEarning), a principled decomposition where Memory handles storage and retrieval infrastructure; Learning extracts intelligence from accumulated interactions asynchronously; and Personalization applies learned knowledge in real-time within finite context budgets. Each component operates as a dedicated sub-agent with specialized tooling and well-defined interfaces. Experimental evaluation on the MAPLE-Personas benchmark demonstrates that our decomposition achieves a 14.6% improvement in personalization score compared to a stateless baseline (p < 0.01, Cohen's d = 0.95) and increases trait incorporation rate from 45% to 75% -- enabling agents that genuinely learn and adapt.
Paper Structure (37 sections, 2 equations, 5 figures, 1 table)

This paper contains 37 sections, 2 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. The Three Capabilities
  3. A Running Example
  4. Background and Related Work
  5. Memory Architectures for LLM Agents
  6. Cognitive Architectures
  7. Personalization in Language Models
  8. The MAPLE Framework
  9. Memory ($\mathcal{M}$): Storage Infrastructure
  10. Three Lenses for Classifying Memory
  11. What Learning Extracts
  12. Levels of Learning
  13. Asynchronous Operation
  14. Personalization ($\mathcal{P}$): Real-Time Adaptation
  15. Levels of Personalization
  16. ...and 22 more sections

Figures (5)

  • Figure 1: Personalization in action: Sarah (senior ML engineer) and Marcus (product manager) ask the same question about transformers. MAPLE retrieves different user profiles and generates tailored responses---technical implementation details for Sarah, conceptual analogies for Marcus.
  • Figure 2: The evolution from foundation models to personalized agents. Base LLMs provide general capabilities; RAG adds knowledge retrieval; MAPLE introduces memory, learning, and personalization as distinct architectural components, enabling agents that adapt to individual users over time.
  • Figure 3: MAPLE architecture as a sequence diagram showing the request-time flow (steps 1--4) and background learning loop (steps 5--6).
  • Figure 4: Context window budget allocation. The finite window must accommodate system prompts (10%), conversation history (20%), tool declarations (10%), user preferences from memory (15%), the current query (5%), and unused context for reasoning (40%). Every token spent on personalization is unavailable for other components.
  • Figure 5: MAPLE vs Baseline across all metrics. Left axis: percentage metrics (trait incorporation, perfect scores). Right axis: mean judge score (1--5 scale). MAPLE achieves statistically significant improvements across all measures (Cohen's $d = 0.95$, $p < 0.01$).