º£½ÇÉçÇø

Host: Prof. Guoning Chen

Ìý

Title: Inspection of Neural Code Intelligence Models

Ìý

Abstract:

Deep neural networks are increasingly being used in various code intelligence tasks such as code summarization, vulnerability detection, type annotation, and many more. While the performance of neural models for intelligent code analysis continues to improve, our understanding of how reliable these models are on unseen data, what features they learn during training, and what are the impacts of noise in training those models are largely unknown. For safe use of such models, researchers often need to reason about the behavior of the underlying models and the factors that affect them. However, this becomes very challenging, as these are opaque models that usually rely on noise-prone data sources for training. This talk presents the latest results of our research in inspection of the neural code intelligence models in terms of generalizability, interpretability, and memorization.