Topological safeguard for evasion attack interpreting the neural networks' behavior

In the last years, Deep Learning technology has been proposed in different fields, bringing many advances in each of them, but identifying new threats in these solutions regarding cybersecurity. Those implemented models have brought several vulnerabilities associated with Deep Learning technology. Moreover, those allow taking advantage of the implemented model, obtaining private information, and even modifying the model's decision-making. Therefore, interest in studying those vulnerabilities/attacks and designing defenses to avoid or fight them is gaining prominence among researchers. In particular, the widely known evasion attack is being analyzed by researchers; thus, several defenses to avoid such a threat can be found in the literature. Since the presentation of the L-BFG algorithm, this threat concerns the research community. However, it continues developing new and ingenious countermeasures since there is no perfect defense for all the known evasion algorithms. In this work, a novel detector of evasion attacks is developed. It focuses on the information of the activations of the neurons given by the model when an input sample is injected. Moreover, it puts attention to the topology of the targeted deep learning model to analyze the activations according to which neurons are connecting. This approach has been decided because the literature shows that the targeted model's topology contains essential information about if the evasion attack occurs. For this purpose, a huge data preprocessing is required to introduce all this information in the detector, which uses the Graph Convolutional Neural Network (GCN) technology. Thus, it understands the topology of the target model, obtaining promising results and improving the outcomes presented in the literature related to similar defenses.

Understanding Deep Learning defenses Against Adversarial Examples Through Visualizations for Dynamic Risk Assessment

In recent years, Deep Neural Network models have been developed in different fields, where they have brought many advances. However, they have also started to be used in tasks where risk is critical. A misdiagnosis of these models can lead to serious accidents or even death. This concern has led to an interest among researchers to study possible attacks on these models, discovering a long list of vulnerabilities, from which every model should be defended. The adversarial example attack is a widely known attack among researchers, who have developed several defenses to avoid such a threat. However, these defenses are as opaque as a deep neural network model, how they work is still unknown. This is why visualizing how they change the behavior of the target model is interesting in order to understand more precisely how the performance of the defended model is being modified. For this work, some defenses, against adversarial example attack, have been selected in order to visualize the behavior modification of each of them in the defended model. Adversarial training, dimensionality reduction and prediction similarity were the selected defenses, which have been developed using a model composed by convolution neural network layers and dense neural network layers. In each defense, the behavior of the original model has been compared with the behavior of the defended model, representing the target model by a graph in a visualization.

Deep Learning Defenses Against Adversarial Examples for Dynamic Risk Assessment

Deep Neural Networks were first developed decades ago, but it was not until recently that they started being extensively used, due to their computing power requirements. Since then, they are increasingly being applied to many fields and have undergone far-reaching advancements. More importantly, they have been utilized for critical matters, such as making decisions in healthcare procedures or autonomous driving, where risk management is crucial. Any mistakes in the diagnostics or decision-making in these fields could entail grave accidents, and even death. This is preoccupying, because it has been repeatedly reported that it is straightforward to attack this type of models. Thus, these attacks must be studied to be able to assess their risk, and defenses need to be developed to make models more robust. For this work, the most widely known attack was selected (adversarial attack) and several defenses were implemented against it (i.e. adversarial training, dimensionality reduc tion and prediction similarity). The obtained outcomes make the model more robust while keeping a similar accuracy. The idea was developed using a breast cancer dataset and a VGG16 and dense neural network model, but the solutions could be applied to datasets from other areas and different convolutional and dense deep neural network models.