Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CaBRNet, an open-source library for developing and evaluating Case-Based Reasoning Models

Romain Xu-Darme, Aymeric Varasse, Alban Grastien, Julien Girard, Zakaria Chihani

TL;DR

CaBRNet is proposed, an open-source, modular, backward-compatible framework for Case-Based Reasoning Networks, an open-source, modular, backward-compatible framework for Case-Based Reasoning Networks.

Abstract

In the field of explainable AI, a vibrant effort is dedicated to the design of self-explainable models, as a more principled alternative to post-hoc methods that attempt to explain the decisions after a model opaquely makes them. However, this productive line of research suffers from common downsides: lack of reproducibility, unfeasible comparison, diverging standards. In this paper, we propose CaBRNet, an open-source, modular, backward-compatible framework for Case-Based Reasoning Networks: https://github.com/aiser-team/cabrnet.

CaBRNet, an open-source library for developing and evaluating Case-Based Reasoning Models

TL;DR

CaBRNet is proposed, an open-source, modular, backward-compatible framework for Case-Based Reasoning Networks, an open-source, modular, backward-compatible framework for Case-Based Reasoning Networks.

Abstract

In the field of explainable AI, a vibrant effort is dedicated to the design of self-explainable models, as a more principled alternative to post-hoc methods that attempt to explain the decisions after a model opaquely makes them. However, this productive line of research suffers from common downsides: lack of reproducibility, unfeasible comparison, diverging standards. In this paper, we propose CaBRNet, an open-source, modular, backward-compatible framework for Case-Based Reasoning Networks: https://github.com/aiser-team/cabrnet.
Paper Structure (9 sections, 2 figures, 1 table)

This paper contains 9 sections, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. CaBRNet: One framework to rule them all...
  3. Modularity
  4. Backward compatibility
  5. Reproducibility and transparency
  6. ... and in the light evaluate them
  7. Going beyond accuracy
  8. Stop wasting time retraining state-of-the-art models
  9. Conclusion and future works

Figures (2)

  • Figure 1: Striving for modularity. From a YAML configuration file (left), CaBRNet backend exploits the common architecture of CBR image classifiers to simplify the instantiation of new models (here, a ProtoTree).
  • Figure 2: Improvement over the perturbation metric used in nauta2020this. Rather than studying the drop in similarity score following a global perturbation of the image (e.g., shift in the hue of the image), we apply a local perturbation using the heatmap produced by the chosen attribution technique. Hence, we not only measure the sensitivity of the similarity score to a given perturbation, but also the ability of the attribution method to locate the most relevant pixels. Additionally, we measure the drop in similarity when applying the perturbation to anything but the most important pixels (dual perturbation).