Attribution-Based Confidence (ABC) Metric

This chapter covers results in the paper **Attribution-Based Confidence Metric For Deep Neural Networks ** available at this author's link 2019 version.

Lecture

The lecture slides are available here.

Introduction

--- primaryColor: steelblue shuffleQuestions: false shuffleAnswers: true --- ### The ABC metric is based on - [ ] the theory of counterfactuals. - [x] the System 1 - System 2 theory. > Correct. - [ ] has no theoretical basis. ### The ABC metric relies mainly on 1. [x] attributions. > Correct. 1. [ ] softmax values of neural networks. 1. [ ] neural network logits. 1. [ ] inputs to the neural network.

ABC Algorithm

--- primaryColor: steelblue shuffleQuestions: false shuffleAnswers: true --- ### ABC metric samples the: - [ ] randomly in the input space. - [x] attribution-neighborhood of the input space. > Correct. - [ ] weights of the neural network.

Results and Conclusions

--- primaryColor: steelblue shuffleQuestions: false shuffleAnswers: true --- ### In this paper, the ABC metric has been applied to - [ ] images and natural language inputs. - [x] images only. > Correct. - [ ] natural language inputs only.

Code and Assignment

There are no programming assignments for this lecture.