Скачать или смотреть Case-Based Reasoning for LLM Agents (April 2025)

Title: Review of Case-Based Reasoning for LLM Agents: Theoretical Foundations, Architectural Components, and Cognitive Integration (Apr 2025)
Link: http://arxiv.org/abs/2504.06943v2
Date: April 2025

Summary:
This paper explores the integration of Case-Based Reasoning (CBR) into LLM agent frameworks to enhance their reasoning and cognitive skills. It reviews theoretical foundations, identifies components, formulates a mathematical model for CBR processes, and evaluates CBR-enhanced agents against other methods. The paper also explores cognitive dimensions and goal-driven autonomy mechanisms to improve LLM agent capabilities.

Key Topics:
Case-Based Reasoning (CBR)
Large Language Models (LLMs)
Retrieval-Augmented Generation (RAG)
Chain-of-Thought Reasoning
Autonomous Agents
Goal-Driven Autonomy
Cognitive Architectures
Meta-cognition

Chapters:
00:00 - Introduction
00:06 - Paper Overview
00:14 - Technically Savvy Listeners
00:26 - Quick Summary
00:38 - Paper Deep Dive
01:01 - Core Issues
01:15 - Hallucinations & Memory
01:27 - Integrating CBR
01:38 - Unpacking the Theory
01:58 - What is Case-Based Reasoning?
02:15 - Core Idea
02:35 - The Standard CBR Cycle
03:17 - Continuous Learning Loop
03:35 - CBR Success
03:57 - Main Benefits
04:14 - Transparency
04:34 - LLM Agents Evolve
04:50 - Perceive, Plan, Act
05:10 - Planning Approaches
05:25 - Memory Challenge
05:39 - Retrieval Augmented Generation
05:56 - Capable Shortcomings
06:06 - Persistent Issues
06:22 - Lack of Memory
06:33 - Synergy of CBR + LLMs
07:00 - Specific Frameworks
07:26 - Healthcare & Law
07:48 - Agents Process
08:06 - Code Generation
08:44 - Cognitive Dimensions
09:17 - Gold-Driven Autonomy
10:14 - Complex Situations
10:44 - Formalizing The Theory
11:24 - Adaptation
11:52 - Case Learning
12:21 - The Architecture
13:04 - Adaptation Strategy
13:36 - The LLM's Reasoning
14:05 - Cognitive Dimensions Continued
14:24 - Introspection
15:10 - Curiosity
15:44 - Richter's Knowledge
16:18 - Goal-Driven Autonomy Again
17:25 - Adaptive
18:17 - Key Question
18:47 - CBR Shine
19:23 - New Domains
19:53 - Efficiency
20:45 - Quality Solutions
21:10 - Theoretical Implications
22:00 - Hurdles
22:21 - What's Next?
23:11 - Rich Research