[A-2] Information Theory

11:00 AM - 11:15 AM JST(2:00 AM - 2:15 AM UTC)Candidate for Academic Incentive Award

[A-2-09] Loss-Based Neuron Path Attribution for Time Series Prediction

◎Ayumi Seki¹, Akihiro Kimura¹, Satoshi Nishiyama¹, Ken Takahashi¹ (1. NTT Inc.)

Keywords:

XAI,Attribution Methods,Machine learning

In recent years, the application of AI in network (NW) operations has advanced significantly. However, when AI outputs are erroneous, methods for analyzing the underlying causes remain insufficiently established, posing a major obstacle to practical deployment. In this paper, we propose a method for quantifying the contribution of individual components to prediction errors in both the training and inference processes of time-series models for network analysis. The proposed method enables the identification of overfitted layers, detection of data bias, and provides insights useful for improving model architectures.