Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DCNFIS: Deep Convolutional Neuro-Fuzzy Inference System

Mojtaba Yeganejou, Kimia Honari, Ryan Kluzinski, Scott Dick, Michael Lipsett, James Miller

TL;DR

DCNFIS addresses interpretability in deep learning by replacing the CNN classifier head with a modified ANFIS, enabling end-to-end training and fuzzy-rule–based explanations. It achieves accuracy on MNIST, Fashion-MNIST, CIFAR-10/100 comparable to the base CNNs and scales to ImageNet with notable gains in interpretability via medoid-based saliency maps, while maintaining competitive performance with state-of-the-art fuzzy methods, as evidenced by $O(N_C \cdot N_V)$ classifier complexity. The approach provides practical explanation mechanisms grounded in fuzzy rules and saliency, enabling debugging and analysis on datasets like Fashion-MNIST and potential counterfactual reasoning, with demonstrated improvements in global class separation in latent spaces. Collectively, the work advances transparent, accurate deep learning by integrating fuzzy inference into CNN backbones and lays groundwork for broader application in large-scale vision tasks and explainable AI research.

Abstract

A key challenge in eXplainable Artificial Intelligence is the well-known tradeoff between the transparency of an algorithm (i.e., how easily a human can directly understand the algorithm, as opposed to receiving a post-hoc explanation), and its accuracy. We report on the design of a new deep network that achieves improved transparency without sacrificing accuracy. We design a deep convolutional neuro-fuzzy inference system (DCNFIS) by hybridizing fuzzy logic and deep learning models and show that DCNFIS performs as accurately as existing convolutional neural networks on four well-known datasets and 3 famous architectures. Our performance comparison with available fuzzy methods show that DCNFIS is now state-of-the-art fuzzy system and outperforms other shallow and deep fuzzy methods to the best of our knowledge. At the end, we exploit the transparency of fuzzy logic by deriving explanations, in the form of saliency maps, from the fuzzy rules encoded in the network to take benefit of fuzzy logic upon regular deep learning methods. We investigate the properties of these explanations in greater depth using the Fashion-MNIST dataset.

DCNFIS: Deep Convolutional Neuro-Fuzzy Inference System

TL;DR

DCNFIS addresses interpretability in deep learning by replacing the CNN classifier head with a modified ANFIS, enabling end-to-end training and fuzzy-rule–based explanations. It achieves accuracy on MNIST, Fashion-MNIST, CIFAR-10/100 comparable to the base CNNs and scales to ImageNet with notable gains in interpretability via medoid-based saliency maps, while maintaining competitive performance with state-of-the-art fuzzy methods, as evidenced by classifier complexity. The approach provides practical explanation mechanisms grounded in fuzzy rules and saliency, enabling debugging and analysis on datasets like Fashion-MNIST and potential counterfactual reasoning, with demonstrated improvements in global class separation in latent spaces. Collectively, the work advances transparent, accurate deep learning by integrating fuzzy inference into CNN backbones and lays groundwork for broader application in large-scale vision tasks and explainable AI research.

Abstract

A key challenge in eXplainable Artificial Intelligence is the well-known tradeoff between the transparency of an algorithm (i.e., how easily a human can directly understand the algorithm, as opposed to receiving a post-hoc explanation), and its accuracy. We report on the design of a new deep network that achieves improved transparency without sacrificing accuracy. We design a deep convolutional neuro-fuzzy inference system (DCNFIS) by hybridizing fuzzy logic and deep learning models and show that DCNFIS performs as accurately as existing convolutional neural networks on four well-known datasets and 3 famous architectures. Our performance comparison with available fuzzy methods show that DCNFIS is now state-of-the-art fuzzy system and outperforms other shallow and deep fuzzy methods to the best of our knowledge. At the end, we exploit the transparency of fuzzy logic by deriving explanations, in the form of saliency maps, from the fuzzy rules encoded in the network to take benefit of fuzzy logic upon regular deep learning methods. We investigate the properties of these explanations in greater depth using the Fashion-MNIST dataset.
Paper Structure (23 sections, 8 equations, 12 figures, 6 tables, 2 algorithms)

This paper contains 23 sections, 8 equations, 12 figures, 6 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Background Review
  3. Fuzzy Classifiers
  4. Convolutional Neural Networks
  5. Deep learning and fuzzy systems
  6. Trust and XAI
  7. Methodology
  8. Proposed Architecture
  9. Learning Algorithm
  10. Guided Backpropagation
  11. Evaluation on Benchmark Datasets
  12. Datasets
  13. Experimental Setup
  14. Evaluation of Results
  15. Evaluation on ImageNet
  16. ...and 8 more sections

Figures (12)

  • Figure 1: Simplified Block Diagram of our Deep Convolutional Neuro-Fuzzy Classifier
  • Figure 2: A simple visualization on guided backpropagation
  • Figure 3: Class representatives derived from fuzzy rules. Top row: MNIST. Middle row: Fashion-MNIST. Bottom row: CIFAR-10
  • Figure 4: Class representatives derived from fuzzy rules for Fashion-MNIST dataset. First row: Medoids. Second row: Saliency of Medoids
  • Figure 5: Difference between T-Shirt and Dress
  • ...and 7 more figures