This book introduces the theory and applications of metaheuristic algorithms. It also provides methods for solving practical problems in such fields as software engineering, image recognition, video networks, and in the oceans. In the theoretical section, the book introduces the information feedback model, learning-based intelligent optimization, dynamic multi-objective optimization, and multi-model optimization. In the applications section, the book presents applications of optimization algorithms to neural architecture search, fuzz testing, oceans, and image processing. The neural architecture search chapter introduces the latest NAS method. The fuzz testing chapter uses multi-objective optimization and ant colony optimization to solve the seed selection and energy allocation problems in fuzz testing. In the ocean chapter, deep learning methods such as CNN, transformer, and attention-based methods are used to describe ENSO prediction and image processing for marine fish identification, and to provide an overview of traditional classification methods and deep learning methods. Rich in examples, this book will be a great resource for students, scholars, and those interested in metaheuristic algorithms, as well as professional practitioners and researchers working on related topics. <!----> <!----> <!----> <!---->
Coverage-based Greybox Fuzzing (CGF) is a practical and effective solution for finding bugs and vulnerabilities in software. A key challenge of CGF is how to select conducive seeds and allocate accurate energy. To address this problem, we propose a novel many-objective optimization solution, MooFuzz, which can identify different states of the seed pool and continuously gather different information about seeds to guide seed schedule and energy allocation. First, MooFuzz conducts risk marking in dangerous positions of the source code. Second, it can automatically update the collected information, including the path risk, the path frequency, and the mutation information. Next, MooFuzz classifies seed pool into three states and adopts different objectives to select seeds. Finally, we design an energy recovery mechanism to monitor energy usage in the fuzzing process and reduce energy consumption. We implement our fuzzing framework and evaluate it on seven real-world programs. The experimental results show that MooFuzz outperforms other state-of-the-art fuzzers, including AFL, AFLFast, FairFuzz, and PerfFuzz, in terms of path discovery and bug detection.
Fuzzing is an extensively used automated vulnerability detection technique. Most existing fuzzers are guided by edge coverage, which makes them less effective in detecting specific vulnerabilities, especially use-after-free (UAF) vulnerabilities. This is because the triggering of a UAF vulnerability must not only cover a specific memory operation but also satisfy a specific sequence of operations. In this paper, we propose UAF-Fuzzer for detecting UAFs, which consists of static analysis and fuzzing stages. In the static analysis stage, UAF-Fuzzer first uses target identification to determine the basic blocks that may cause UAFs as the target basic blocks; subsequently, it then instruments these target basic blocks. Subsequently, we propose a memory operation evaluation method to assess the complexity of memory operations. In the fuzzing stage, UAF-Fuzzer assigns energy to seeds using a memory evaluation operation and employs a novel seed selection algorithm to prioritize the execution of test cases that are likely to trigger UAF vulnerabilities. We designed and implemented a UAF-Fuzzer to improve the detection of UAFs and compared it with AFL, AFLFast, FairFuzz, MOPT, EcoFuzz, and TortoiseFuzz in terms of UAF vulnerability detection, crash detection, and path discovery. The results showed that UAF-Fuzzer is more effective in terms of detecting UAF vulnerabilities. We have also discovered three UAF vulnerabilities, submitted them to the software maintainer for fixing, and obtained CVE IDs.
Evolutionary algorithms play an irreplaceable role in the optimization of deep neural network architecture. In this chapter, the basic definition of neural architecture search (NAS) is given first; then the search strategy and architecture space of the NAS approaches are invested in detailed. To jointly optimize the CNN unit architecture and depth, an automatic architecture design method is introduced that condenses the CNN architecture space into architecture representation and then uses monarch butterfly optimization (MBO) to search the representation to obtain available architectures. Finally, a block-based NAS method is introduced that conducts a CNN architecture search based on ResNet block and DenseNet block, which achieved competitive results on two important image benchmarks. <!----> <!----> <!----> <!---->
Multimodal multi-objective optimization has become a popular research field in recent years. It has similar features to the multi-objective optimization, e.g., optimizing multiple objectives simultaneously and eventually finding global optima. Nevertheless, multimodal multi-objective optimization has several equivalent Pareto optimal solution sets, differing from one another and corresponding to the same Pareto optimal front, which is different from the Pareto optimal solution sets corresponding to one Pareto optimal front in multi-objective optimization. To solve multimodal multi-objective optimization problems properly, adjustments should be performed in the traditional approaches, i.e., multi-objective evolutionary algorithms, in solving multi-objective optimization problems, since they perform poorly when directly utilized in multimodal multi-objective optimization problems. A great deal of research and work about multimodal multi-objective optimization problems are provided in detail in Chapter 5. <!----> <!----> <!----> <!---->