Neural Network Representation Control: Gaussian Isolation Machines and CVC Regularization

In many cases, neural network classifiers are likely to be exposed to input data that is outside of their training distribution data. Samples from outside the distribution may be classified as an existing class with high probability by softmax-based classifiers; such incorrect classifications affect the performance of the classifiers and the applications/systems that depend on them. Previous research aimed at distinguishing training distribution data from out-of-distribution data (OOD) has proposed detectors that are external to the classification method. We present Gaussian isolation machine (GIM), a novel hybrid (generative-discriminative) classifier aimed at solving the problem arising when OOD data is encountered. The GIM is based on a neural network and utilizes a new loss function that imposes a distribution on each of the trained classes in the neural network's output space, which can be approximated by a Gaussian. The proposed GIM's novelty lies in its discriminative performance and generative capabilities, a combination of characteristics not usually seen in a single classifier. The GIM achieves state-of-the-art classification results on image recognition and sentiment analysis benchmarking datasets and can also deal with OOD inputs. We also demonstrate the benefits of incorporating part of the GIM's loss function into standard neural networks as a regularization method.

Exploring the Back Alleys: Analysing The Robustness of Alternative Neural Network Architectures against Adversarial Attacks

Recent discoveries in the field of adversarial machine learning have shown that Artificial Neural Networks (ANNs) are susceptible to adversarial attacks. These attacks cause misclassification of specially crafted adversarial samples. In light of this phenomenon, it is worth investigating whether other types of neural networks are less susceptible to adversarial attacks. In this work, we applied standard attack methods originally aimed at conventional ANNs, towards stochastic ANNs and also towards Spiking Neural Networks (SNNs), across three different datasets namely MNIST, CIFAR-10 and Patch Camelyon. We analysed their adversarial robustness against attacks performed in the raw image space of the different model variants. We employ a variety of attacks namely Basic Iterative Method (BIM), Carlini & Wagner L2 attack (CWL2) and Boundary attack. Our results suggests that SNNs and stochastic ANNs exhibit some degree of adversarial robustness as compared to their ANN counterparts under certain attack methods. Namely, we found that the Boundary and the state-of-the-art CWL2 attacks are largely ineffective against stochastic ANNs. Following this observation, we proposed a modified version of the CWL2 attack and analysed the impact of this attack on the models' adversarial robustness. Our results suggest that with this modified CWL2 attack, many models are more easily fooled as compared to the vanilla CWL2 attack, albeit observing an increase in L2 norms of adversarial perturbations. Lastly, we also investigate the resilience of alternative neural networks against adversarial samples transferred from ResNet18. We show that the modified CWL2 attack provides an improved cross-architecture transferability compared to other attacks.

TEAGS: Time-aware Text Embedding Approach to Generate Subgraphs

Contagions (e.g. virus, gossip) spread over the nodes in propagation graphs. We can use the temporal and textual data of the nodes to compute the edge weights and then generate subgraphs with highly relevant nodes. This is beneficial to many applications. Yet, challenges abound. First, the propagation pattern between each pair of nodes may change by time. Second, not always the same contagion propagates. Hence, the state-of-the-art text mining approaches including topic-modeling cannot effectively compute the edge weights. Third, since the propagation is affected by time, the word-word co-occurrence patterns may differ in various temporal dimensions, that can decrease the effectiveness of word embedding approaches. We argue that multi-aspect temporal dimensions (hour, day, etc) should be considered to better calculate the correlation weights between the nodes. In this work, we devise a novel framework that on the one hand, integrates a neural network based time-aware word embedding component to construct the word vectors through multiple temporal facets, and on the other hand, uses a temporal generative model to compute the weights. Subsequently, we propose a Max-Heap Graph cutting algorithm to generate subgraphs. We validate our model through comprehensive experiments on real-world datasets. The results show that our model can retrieve the subgraphs more effective than other rivals and the temporal dynamics should be noticed both in word embedding and propagation processes.

Why Blocking Targeted Adversarial Perturbations Impairs the Ability to Learn

Despite their accuracy, neural network-based classifiers are still prone to manipulation through adversarial perturbations. Those perturbations are designed to be misclassified by the neural network, while being perceptually identical to some valid input. The vast majority of attack methods rely on white-box conditions, where the attacker has full knowledge of the attacked network's parameters. This allows the attacker to calculate the network's loss gradient with respect to some valid input and use this gradient in order to create an adversarial example. The task of blocking white-box attacks has proven difficult to solve. While a large number of defense methods have been suggested, they have had limited success. In this work we examine this difficulty and try to understand it. We systematically explore the abilities and limitations of defensive distillation, one of the most promising defense mechanisms against adversarial perturbations suggested so far in order to understand the defense challenge. We show that contrary to commonly held belief, the ability to bypass defensive distillation is not dependent on an attack's level of sophistication. In fact, simple approaches, such as the Targeted Gradient Sign Method, are capable of effectively bypassing defensive distillation. We prove that defensive distillation is highly effective against non-targeted attacks but is unsuitable for targeted attacks. This discovery leads us to realize that targeted attacks leverage the same input gradient that allows a network to be trained. This implies that blocking them will require losing the network's ability to learn, presenting an impossible tradeoff to the research community.

Privacy-Preserving Detection of IoT Devices Connected Behind a NAT in a Smart Home Setup

Today, telecommunication service providers (telcos) are exposed to cyber-attacks executed by compromised IoT devices connected to their customers' networks. Such attacks might have severe effects not only on the target of attacks but also on the telcos themselves. To mitigate those risks we propose a machine learning based method that can detect devices of specific vulnerable IoT models connected behind a domestic NAT, thereby identifying home networks that pose a risk to the telco's infrastructure and availability of services. As part of the effort to preserve the domestic customers' privacy, our method relies on NetFlow data solely, refraining from inspecting the payload. To promote future research in this domain we share our novel dataset, collected in our lab from numerous and various commercial IoT devices.

Adversarial Attacks on Remote User Authentication Using Behavioural Mouse Dynamics

Mouse dynamics is a potential means of authenticating users. Typically, the authentication process is based on classical machine learning techniques, but recently, deep learning techniques have been introduced for this purpose. Although prior research has demonstrated how machine learning and deep learning algorithms can be bypassed by carefully crafted adversarial samples, there has been very little research performed on the topic of behavioural biometrics in the adversarial domain. In an attempt to address this gap, we built a set of attacks, which are applications of several generative approaches, to construct adversarial mouse trajectories that bypass authentication models. These generated mouse sequences will serve as the adversarial samples in the context of our experiments. We also present an analysis of the attack approaches we explored, explaining their limitations. In contrast to previous work, we consider the attacks in a more realistic and challenging setting in which an attacker has access to recorded user data but does not have access to the authentication model or its outputs. We explore three different attack strategies: 1) statistics-based, 2) imitation-based, and 3) surrogate-based; we show that they are able to evade the functionality of the authentication models, thereby impacting their robustness adversely. We show that imitation-based attacks often perform better than surrogate-based attacks, unless, however, the attacker can guess the architecture of the authentication model. In such cases, we propose a potential detection mechanism against surrogate-based attacks.

CT-GAN: Malicious Tampering of 3D Medical Imagery using Deep Learning

In 2018, clinics and hospitals were hit with numerous attacks leading to significant data breaches and interruptions in medical services. An attacker with access to medical records can do much more than hold the data for ransom or sell it on the black market. In this paper, we show how an attacker can use deep learning to add or remove evidence of medical conditions from volumetric (3D) medical scans. An attacker may perform this act in order to stop a political candidate, sabotage research, commit insurance fraud, perform an act of terrorism, or even commit murder. We implement the attack using a 3D conditional GAN and show how the framework (CT-GAN) can be automated. Although the body is complex and 3D medical scans are very large, CT-GAN achieves realistic results and can be executed in milliseconds. To evaluate the attack, we focus on injecting and removing lung cancer from CT scans. We show how three expert radiologists and a state-of-the-art deep learning AI could not differentiate between tampered and non-tampered scans. We also evaluate state-of-the-art countermeasures and propose our own. Finally, we discuss the possible attack vectors on modern radiology networks and demonstrate one of the attack vectors on an active CT scanner.

Detecting Adversarial Perturbations Through Spatial Behavior in Activation Spaces

Neural network based classifiers are still prone to manipulation through adversarial perturbations. State of the art attacks can overcome most of the defense or detection mechanisms suggested so far, and adversaries have the upper hand in this arms race. Adversarial examples are designed to resemble the normal input from which they were constructed, while triggering an incorrect classification. This basic design goal leads to a characteristic spatial behavior within the context of Activation Spaces, a term coined by the authors to refer to the hyperspaces formed by the activation values of the network's layers. Within the output of the first layers of the network, an adversarial example is likely to resemble normal instances of the source class, while in the final layers such examples will diverge towards the adversary's target class. The steps below enable us to leverage this inherent shift from one class to another in order to form a novel adversarial example detector. We construct Euclidian spaces out of the activation values of each of the deep neural network layers. Then, we induce a set of k-nearest neighbor classifiers (k-NN), one per activation space of each neural network layer, using the non-adversarial examples. We leverage those classifiers to produce a sequence of class labels for each nonperturbed input sample and estimate the a priori probability for a class label change between one activation space and another. During the detection phase we compute a sequence of classification labels for each input using the trained classifiers. We then estimate the likelihood of those classification sequences and show that adversarial sequences are far less likely than normal ones. We evaluated our detection method against the state of the art C&W attack method, using two image classification datasets (MNIST, CIFAR-10) reaching an AUC 0f 0.95 for the CIFAR-10 dataset.

Query-Efficient GAN Based Black-Box Attack Against Sequence Based Machine and Deep Learning Classifiers

In this paper, we present a generic black-box attack, demonstrated against API call based machine learning malware classifiers. We generate adversarial examples combining sequences (API call sequences) and other features (e.g., printable strings) that will be misclassified by the classifier without affecting the malware functionality. Our attack minimizes the number of target classifier queries and only requires access to the predicted label of the attacked model (without the confidence level). We evaluate the attack's effectiveness against many classifiers such as RNN variants, DNN, SVM, GBDT, etc. We show that the attack requires fewer queries and less knowledge about the attacked model's architecture than other existing black-box attacks, making it optimal to attack cloud based models at a minimal cost. Finally, we discuss the robustness of this attack to existing defense mechanisms.

DOPING: Generative Data Augmentation for Unsupervised Anomaly Detection with GAN

Recently, the introduction of the generative adversarial network (GAN) and its variants has enabled the generation of realistic synthetic samples, which has been used for enlarging training sets. Previous work primarily focused on data augmentation for semi-supervised and supervised tasks. In this paper, we instead focus on unsupervised anomaly detection and propose a novel generative data augmentation framework optimized for this task. In particular, we propose to oversample infrequent normal samples - normal samples that occur with small probability, e.g., rare normal events. We show that these samples are responsible for false positives in anomaly detection. However, oversampling of infrequent normal samples is challenging for real-world high-dimensional data with multimodal distributions. To address this challenge, we propose to use a GAN variant known as the adversarial autoencoder (AAE) to transform the high-dimensional multimodal data distributions into low-dimensional unimodal latent distributions with well-defined tail probability. Then, we systematically oversample at the `edge' of the latent distributions to increase the density of infrequent normal samples. We show that our oversampling pipeline is a unified one: it is generally applicable to datasets with different complex data distributions. To the best of our knowledge, our method is the first data augmentation technique focused on improving performance in unsupervised anomaly detection. We validate our method by demonstrating consistent improvements across several real-world datasets.