No abstract

Abstract-Researchers have proposed formal definitions of quantitative information flow based on information theoretic notions such as the Shannon entropy, the min entropy, the guessing entropy, and channel capacity. This paper investigates the hardness and possibilities of precisely checking and inferring quantitative information flow according to such definitions.We prove that, even for just comparing two programs on which has the larger flow, none of the definitions is a ksafety property for any k, and therefore is not amenable to the self-composition technique that has been successfully applied to precisely checking non-interference. We also show a complexity theoretic gap with non-interference by proving that, for loop-free boolean programs whose non-interference is coNP-complete, the comparison problem is #P-hard for all of the definitions.For positive results, we show that universally quantifying the distribution in the comparison problem, that is, comparing two programs according to the entropy based definitions on which has the larger flow for all distributions, is a 2-safety problem in general and is coNP-complete when restricted for loop-free boolean programs. We prove this by showing that the problem is equivalent to a simple relation naturally expressing the fact that one program is more secure than the other. We prove that the relation also refines the channel-capacity based definition, and that it can be precisely checked via the self-composition as well as the "interleaved" self-composition technique.

For using neural networks in safety critical domains, it is important to know if a decision made by a neural network is supported by prior similarities in training. We propose runtime neuron activation pattern monitoring -after the standard training process, one creates a monitor by feeding the training data to the network again in order to store the neuron activation patterns in abstract form. In operation, a classification decision over an input is further supplemented by examining if a pattern similar (measured by Hamming distance) to the generated pattern is contained in the monitor. If the monitor does not contain any pattern similar to the generated pattern, it raises a warning that the decision is not based on the training data. Our experiments show that, by adjusting the similarity-threshold for activation patterns, the monitors can report a significant portion of misclassfications to be not supported by training with a small false-positive rate, when evaluated on a test set.

Systematically testing models learned from neural networks remains a crucial unsolved barrier to successfully justify safety for autonomous vehicles engineered using data-driven approach. We propose quantitative k-projection coverage as a metric to mediate combinatorial explosion while guiding the data sampling process. By assuming that domain experts propose largely independent environment conditions and by associating elements in each condition with weights, the product of these conditions forms scenarios, and one may interpret weights associated with each equivalence class as relative importance. Achieving full k-projection coverage requires that the data set, when being projected to the hyperplane formed by arbitrarily selected k-conditions, covers each class with number of data points no less than the associated weight. For the general case where scenario composition is constrained by rules, precisely computing k-projection coverage remains in NP. In terms of finding minimum test cases to achieve full coverage, we present theoretical complexity for important sub-cases and an encoding to 0-1 integer programming. We have implemented a research prototype that generates test cases for a visual object detection unit in automated driving, demonstrating the technological feasibility of our proposed coverage criterion. IBM CPLEX OptimizationStudio: https

Abstract. Researchers have proposed formal definitions of quantitative information flow based on information theoretic notions such as the Shannon entropy, the min entropy, the guessing entropy, and channel capacity. This paper investigates the hardness of precisely checking the quantitative information flow of a program according to such definitions. More precisely, we study the "bounding problem" of quantitative information flow, defined as follows: Given a program M and a positive real number q, decide if the quantitative information flow of M is less than or equal to q. We prove that the bounding problem is not a k-safety property for any k (even when q is fixed, for the Shannon-entropy-based definition with the uniform distribution), and therefore is not amenable to the self-composition technique that has been successfully applied to checking non-interference. We also prove complexity theoretic hardness results for the case when the program is restricted to loop-free boolean programs. Specifically, we show that the problem is PP-hard for all the definitions, showing a gap with non-interference which is coNP-complete for the same class of programs. The paper also compares the results with the recently proved results on the comparison problems of quantitative information flow.