Biblio

Most of the data manipulation attacks on deep neural networks (DNNs) during the training stage introduce a perceptible noise that can be catered by preprocessing during inference, or can be identified during the validation phase. There-fore, data poisoning attacks during inference (e.g., adversarial attacks) are becoming more popular. However, many of them do not consider the imperceptibility factor in their optimization algorithms, and can be detected by correlation and structural similarity analysis, or noticeable (e.g., by humans) in multi-level security system. Moreover, majority of the inference attack rely on some knowledge about the training dataset. In this paper, we propose a novel methodology which automatically generates imperceptible attack images by using the back-propagation algorithm on pre-trained DNNs, without requiring any information about the training dataset (i.e., completely training data-unaware). We present a case study on traffic sign detection using the VGGNet trained on the German Traffic Sign Recognition Benchmarks dataset in an autonomous driving use case. Our results demonstrate that the generated attack images successfully perform misclassification while remaining imperceptible in both “subjective” and “objective” quality tests.

We propose a distributed continuous-time algorithm to solve a network optimization problem where the global cost function is a strictly convex function composed of the sum of the local cost functions of the agents. We establish that our algorithm, when implemented over strongly connected and weight-balanced directed graph topologies, converges exponentially fast when the local cost functions are strongly convex and their gradients are globally Lipschitz. We also characterize the privacy preservation properties of our algorithm and extend the convergence guarantees to the case of time-varying, strongly connected, weight-balanced digraphs. When the network topology is a connected undirected graph, we show that exponential convergence is still preserved if the gradients of the strongly convex local cost functions are locally Lipschitz, while it is asymptotic if the local cost functions are convex. We also study discrete-time communication implementations. Specifically, we provide an upper bound on the stepsize of a synchronous periodic communication scheme that guarantees convergence over connected undirected graph topologies and, building on this result, design a centralized event-triggered implementation that is free of Zeno behavior. Simulations illustrate our results.

We propose a distributed continuous-time algorithm to solve a network optimization problem where the global cost function is a strictly convex function composed of the sum of the local cost functions of the agents. We establish that our algorithm, when implemented over strongly connected and weight-balanced directed graph topologies, converges exponentially fast when the local cost functions are strongly convex and their gradients are globally Lipschitz. We also characterize the privacy preservation properties of our algorithm and extend the convergence guarantees to the case of time-varying, strongly connected, weight-balanced digraphs. When the network topology is a connected undirected graph, we show that exponential convergence is still preserved if the gradients of the strongly convex local cost functions are locally Lipschitz, while it is asymptotic if the local cost functions are convex. We also study discrete-time communication implementations. Specifically, we provide an upper bound on the stepsize of a synchronous periodic communication scheme that guarantees convergence over connected undirected graph topologies and, building on this result, design a centralized event-triggered implementation that is free of Zeno behavior. Simulations illustrate our results.