Baseline Computation for Attribution Methods Based on Interpolated Inputs
arXiv:2204.06120v1h-index: 4
Originality Synthesis-oriented
AI Analysis
This work is incremental, focusing on improving baseline computation for specific attribution methods in machine learning interpretability.
The paper addresses the challenge of selecting a well-behaved baseline for attribution methods that use interpolated inputs, and tests this approach with a new method called RSI-Grad-CAM, though no concrete results or numbers are provided.
We discuss a way to find a well behaved baseline for attribution methods that work by feeding a neural network with a sequence of interpolated inputs between two given inputs. Then, we test it with our novel Riemann-Stieltjes Integrated Gradient-weighted Class Activation Mapping (RSI-Grad-CAM) attribution method.