Retrieval-Augmented Automated Program Repair System

A Retrieval-Augmented Automated Program Repair System is an automated program repair system that implements retrieval-augmented generation techniques to enhance patch generation.

• AKA: RAG-Based Program Repair System, Retrieval-Enhanced Code Repair System, Example-Driven Repair System.
• Context:
  • It can typically retrieve Relevant Bug-Fix Examples from code repositorys.
  • It can typically augment Language Model Inputs with retrieved context.
  • It can typically improve Patch Accuracy through knowledge grounding.
  • It can often utilize Semantic Search for relevant retrieval.
  • It can often employ Vector Databases for efficient indexing.
  • ...
  • It can range from being a Single-Source Retrieval System to being a Multi-Source Retrieval System, depending on its data source diversity.
  • It can range from being a Static Retrieval System to being a Adaptive Retrieval System, depending on its retrieval strategy.
  • It can range from being a Lexical Retrieval System to being a Semantic Retrieval System, depending on its retrieval mechanism.
  • It can range from being a Local Retrieval System to being a Cloud-Based Retrieval System, depending on its deployment architecture.
  • ...
  • It can integrate with Large Language Models for patch synthesis.
  • It can utilize Code Similarity Measures for retrieval ranking.
  • It can employ Information Retrieval Systems for knowledge access.
  • ...
• Example(s):
  • CURE System using neural retrieval for bug-fix examples.
  • RAP-Gen System retrieving relevant patches for repair generation.
  • SelRepair System with dual retrieval mechanisms.
  • Cedar System combining retrieval with program synthesis.
  • RewardRepair System using reinforcement learning for retrieval optimization.
  • Getafix System by Meta for learning-based retrieval.
  • DeepDebug System with multi-modal retrieval.
  • ...
• Counter-Example(s):
  • Template-Based Repair System, which uses static patterns.
  • Pure Neural Repair System, which relies only on model parameters.
  • Search-Based Repair System, which explores solution space without retrieval.
• See: Retrieval-Augmented Natural Language Generation Technique, Automated Program Repair System, Information Retrieval System, Large Language Model.

---