A Hybrid Post Hoc Interpretability Approach for Deep Neural Networks

Flávio Arthur Oliveira Santos; Cleber Zanchettin; José Vitor Santos Silva; Leonardo Nogueira Matos; Paulo Novais

Ayuda

A Hybrid Post Hoc Interpretability Approach for Deep Neural Networks

Santos, Flávio Arthur Oliveira ^[1] ^[3] ; Cleber Zanchettin ^[1] ; Silva, José Vitor Santos ^[2] ; Matos, Leonardo Nogueira ^[2] ; Paulo Novais ^[3]
1. [1] Universidade Federal de Pernambuco
  
  Universidade Federal de Pernambuco
  
  Brasil
2. [2] Universidade Federal de Sergipe
  
  Universidade Federal de Sergipe
  
  Brasil
3. [3] Universidade do Minho
  
  Universidade do Minho
  
  Braga (São José de São Lázaro), Portugal
Mostrar afiliaciones +
Localización: Hybrid Artificial Intelligent Systems: 16th International Conference, HAIS 2021. Bilbao, Spain. September 22–24, 2021. Proceedings / coord. por Hugo Sanjurjo González, Iker Pastor López , Pablo García Bringas , Héctor Quintián Pardo , Emilio Santiago Corchado Rodríguez , 2021, ISBN 978-3-030-86271-8, págs. 600-610
Idioma: inglés
Texto completo no disponible (Saber más ...)
Resumen
- Every day researchers publish works with state-of-the-art results using deep learning models, however as these models become common even in production, ensuring fairness is a main concern of the deep learning models. One way to analyze the model fairness is based on the model interpretability, obtaining the essential features to the model decision. There are many interpretability methods to produce the deep learning model interpretation, such as Saliency, GradCam, Integrated Gradients, Layer-wise relevance propagation, and others. Although those methods make the feature importance map, different methods have different interpretations, and their evaluation relies on qualitative analysis. In this work, we propose the Iterative post hoc attribution approach, which consists of seeing the interpretability problem as an optimization view guided by two objective definitions of what our solution considers important. We solve the optimization problem with a hybrid approach considering the optimization algorithm and the deep neural network model. The obtained results show that our approach can select the features essential to the model prediction more accurately than the traditional interpretability methods.