Belangrijkste concepten
プロジェクト固有のショートカットに依存することで、ニューラルコードモデルが一般化できない問題を解明し、新しいバイアス軽減メカニズムを提案。
Statistieken
Deep learning has introduced significant improvements in many software analysis tasks.
Large Language Models (LLMs) based neural code models demonstrate commendable performance within the intra-project IID setting.
Neural code models often struggle to generalize effectively to real-world inter-project out-of-distribution (OOD) data.
Citaten
"Without proper regularization, models tend to leverage spurious statistical cues for prediction."
"Project-specific bias learning behavior can be interpreted with a proposed measurement called Cond-Idf."