Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Efficient Visualization of Neural Networks with Generative Models and Adversarial Perturbations

Athanasios Karagounis

TL;DR

A novel approach for deep visualization via a generative network, offering an improvement over existing methods that incorporates a unique skip-connection-inspired block design, which enhances label-directed image generation by propagating class information across multiple layers.

Abstract

This paper presents a novel approach for deep visualization via a generative network, offering an improvement over existing methods. Our model simplifies the architecture by reducing the number of networks used, requiring only a generator and a discriminator, as opposed to the multiple networks traditionally involved. Additionally, our model requires less prior training knowledge and uses a non-adversarial training process, where the discriminator acts as a guide rather than a competitor to the generator. The core contribution of this work is its ability to generate detailed visualization images that align with specific class labels. Our model incorporates a unique skip-connection-inspired block design, which enhances label-directed image generation by propagating class information across multiple layers. Furthermore, we explore how these generated visualizations can be utilized as adversarial examples, effectively fooling classification networks with minimal perceptible modifications to the original images. Experimental results demonstrate that our method outperforms traditional adversarial example generation techniques in both targeted and non-targeted attacks, achieving up to a 94.5% fooling rate with minimal perturbation. This work bridges the gap between visualization methods and adversarial examples, proposing that fooling rate could serve as a quantitative measure for evaluating visualization quality. The insights from this study provide a new perspective on the interpretability of neural networks and their vulnerabilities to adversarial attacks.

Efficient Visualization of Neural Networks with Generative Models and Adversarial Perturbations

TL;DR

A novel approach for deep visualization via a generative network, offering an improvement over existing methods that incorporates a unique skip-connection-inspired block design, which enhances label-directed image generation by propagating class information across multiple layers.

Abstract

This paper presents a novel approach for deep visualization via a generative network, offering an improvement over existing methods. Our model simplifies the architecture by reducing the number of networks used, requiring only a generator and a discriminator, as opposed to the multiple networks traditionally involved. Additionally, our model requires less prior training knowledge and uses a non-adversarial training process, where the discriminator acts as a guide rather than a competitor to the generator. The core contribution of this work is its ability to generate detailed visualization images that align with specific class labels. Our model incorporates a unique skip-connection-inspired block design, which enhances label-directed image generation by propagating class information across multiple layers. Furthermore, we explore how these generated visualizations can be utilized as adversarial examples, effectively fooling classification networks with minimal perceptible modifications to the original images. Experimental results demonstrate that our method outperforms traditional adversarial example generation techniques in both targeted and non-targeted attacks, achieving up to a 94.5% fooling rate with minimal perturbation. This work bridges the gap between visualization methods and adversarial examples, proposing that fooling rate could serve as a quantitative measure for evaluating visualization quality. The insights from this study provide a new perspective on the interpretability of neural networks and their vulnerabilities to adversarial attacks.
Paper Structure (9 sections, 3 figures)

This paper contains 9 sections, 3 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. Generative Adversarial Networks
  4. Adversarial Examples
  5. Proposed Method
  6. Network Architecture
  7. Training Process
  8. Experimental Results
  9. Conclusions

Figures (3)

  • Figure 1: The illustration of the proposed model. Bottom row represents the overall structure. The proposed generative network is shown on the left and the classification network that needs to be visualized is shown on the right. Given a class id (red cube) as input, the generator can generate an image. The discriminator makes a prediction on the generated image. The loss is measured by the distance between input class id (red cube on the left) and predicted class id (red circle on the right). We freeze the discriminator’s weights and only train the generator. The top row represent our designed block to encode the input class id information (pink arrow in the bottom row) to layers of different depth for generator network.
  • Figure 2: Adversarial examples and their corresponding mistaken labels.
  • Figure 3: The relationship of multiplied coefficient and fooling rate.