ARTFEED — Contemporary Art Intelligence

HyperFL: Query-Adaptive Representation Learning for Software Fault Localization

ai-technology · 2026-08-06

A novel framework named HyperFL has been introduced to enhance software fault localization by tailoring query representations to the variety found in real-world issue reports. Conventional retrieval methods utilize a static query representation for all issue reports, neglecting differences in length, structure, and debugging details. HyperFL employs a lightweight hypernetwork to create query-specific LoRA parameters for the query encoder, enabling flexible adaptation while maintaining a fixed and reusable code encoder. Tests conducted on a real-world issue localization benchmark demonstrate consistent advancements. This research is documented in arXiv paper 2608.02967, which was announced as a cross-type submission.

Key facts

  • HyperFL is a query-adaptive representation learning framework for software fault localization.
  • It addresses the limitation of fixed query representations in retrieval-based fault localization.
  • A lightweight hypernetwork generates query-specific LoRA parameters for the query encoder.
  • The code encoder remains fixed and reusable.
  • Experiments on a real-world issue localization benchmark show consistent improvements.
  • The paper is available on arXiv with ID 2608.02967.
  • The announcement type is 'cross'.
  • The approach is designed for automated debugging and program repair.

Entities

Institutions

  • arXiv

Sources