The Android ecosystem has recently dominated mobile devices. Android app markets, including official Google Play and other third party markets, are becoming hotbeds, where malware originates and spreads. Android malware has been observed to both propagate within markets and spread between markets. If the spread of Android malware between markets can be predicted, market administrators can take appropriate measures to prevent the outbreak of malware and minimize the damages caused by malware. In this paper, we make the first attempt to protect the Android ecosystem by modeling and predicting the spread of Android malware between markets. To this end, we study the social behaviors that affect the spread of malware, model these spread behaviors with multiple epidemic models, and predict the infection time and order among markets for well-known malware families. To achieve an accurate prediction of malware spread, we model spread behaviors in the following fashion: 1) for a single market, we model the within-market malware growth by considering both the creation and removal of malware; 2) for multiple markets, we determine market relevance by calculating the mutual information among them; and 3) based on the previous two steps, we simulate a susceptible infected model stochastically for spread among markets. The model inference is performed using a publicly available well-labeled dataset AndRadar. To conduct extensive experiments to evaluate our approach, we collected a large number (334,782) of malware samples from 25 Android markets around the world. The experimental results show our approach can depict and simulate the growth of Android malware on a large scale, and predict the infection time and order among markets with 0.89 and 0.66 precision, respectively.

Stochastic simulations are developed and employed across many fields, to advise governmental policy decisions and direct future research. Faulty simulation software can have serious consequences, but its correctness is difficult to determine due to complexity and random behaviour. Stochastic simulations may output a different result each time they are run, whereas most testing techniques are designed for programs which (for a given set of inputs) always produce the same behaviour. In this paper, we introduce a new approach towards testing stochastic simulations using statistical oracles and transition probabilities. Our approach was implemented as a toolkit, which allows the frequency of state transitions to be tested, along with their final output distribution. We evaluated our toolkit on eight simulation programs from a variety fields and found it can detect errors at least three times smaller (and in one case, over 1000 times smaller) than a conventional (tolerance threshold) approach.

Stochastic models can be difficult to test due to their complexity and randomness, yet their predictions are often used to make important decisions, so they need to be correct. We introduce a new search-based technique for testing implementations of stochastic models by maximising the differences between the implementation and a pseudo-oracle. Our technique reduces testing effort and enables discrepancies to be found that might otherwise be overlooked. We show the technique can identify differences challenging for humans to observe, and use it to help a new user understand implementation differences in a real model of a citrus disease (Huanglongbing) used to inform policy and research.

It is difficult to test programs that input images, due to the large number of (pixel) values that must be chosen and the complex ways these values interact. Typically, such programs are tested manually, using images that have known results. However, this is a laborious process and limited in the range of tests that can be applied. We introduce a new approach for testing programs that input images automatically, using procedural noise and spatial statistics to create inputs that are both realistic and can easily be tuned to have specific properties. The effectiveness of our approach is illustrated on an epidemiological simulation of a recently introduced tree pest in Great Britain: Oriental Chestnut Gall Wasp. Our approach produces images that match the real landscapes more closely than other techniques and can be used (alongside metamorphic relations) to detect smaller (artificially introduced) errors with greater accuracy.

We can only test software effectively if we understand how it is intended to behave.For some categories of programs, such as scientific models, it is not obvious what the output of the software should be.New techniques are needed to help domain experts, such as scientists, gather the knowledge they need to construct suitable tests and oracles.This paper introduces a new interactive tool for iterative test suite construction that is based upon the scientific method paradigm that scientists are familiar with.We apply our technique to a deterministic mathematical model, used to predict the spread of disease, and show how it helps scientists uncover situations they had not yet considered.Of the 15 hypotheses originally created by modellers, our technique found discrepancies in all but one, allowing us to refine them into a more rigorous test suite.

Test selection techniques are used to reduce the human effort involved in software testing. Most research focusses on selecting efficient sets of test cases according to various coverage criteria for directed testing. We introduce a new technique to select efficient sets of sub domains from which new test cases can be sampled at random to achieve a high mutation score. We first present a technique for evolving multiple sub domains, each of which target a different group of mutants. The evolved sub domains are shown to achieve an average 160% improvement in mutation score compared to random testing with six real world Java programs. We then present a technique for selecting sets of the evolved sub domains to reduce the human effort involved in evaluating sampled test cases without reducing their fault finding effectiveness. This technique significantly reduces the number of sub domains for four of the six programs with a negligible difference in mutation score.

Mutation testing is effective at measuring the adequacy of a test suite, but it can be computationally expensive to apply all the test cases to each mutant. Previous research has investigated the effect of reducing the number of mutants by selecting certain operators, sampling mutants at random, or combining them to form new higher-order mutants. In this paper, we propose a new approach to the mutant reduction problem using static analysis. Symbolic representations are generated for the output along the paths through each mutant and these are compared with the original program. By calculating the range of their output expressions, it is possible to determine the effect of each mutation on the program output. Mutants with little effect on the output are harder to kill. We confirm this using random testing and an established test suite. Competent programmers are likely to only make small mistakes in their programming code. We argue therefore that test suites should be evaluated against those mutants that are harder to kill without being equivalent to the original program.