Fuzzing is becoming more and more popular in the field of vulnerability detection. In the process of fuzzing, seed selection strategy plays an important role in guiding the evolution direction of fuzzing. However, the SOTA fuzzers only focus on indiv
idual uncertainty, neglecting the multi-factor uncertainty caused by both randomization and evolution. In this paper, we consider seed selection in fuzzing as a large-scale online planning problem under uncertainty. We propose mytool which is a new intelligent seed selection strategy. In Alpha-Fuzz, we leverage the MCTS algorithm to deal with the effects of the uncertainty of randomization and evolution of fuzzing. Especially, we analyze the role of the evolutionary relationship between seeds in the process of fuzzing, and propose a new tree policy and a new default policy to make the MCTS algorithm better adapt to the fuzzing. We compared mytool with four state-of-the-art fuzzers in 12 real-world applications and LAVA-M data set. The experimental results show that mytool could find more bugs on lava-M and outperforms other tools in terms of code coverage and number of bugs discovered in the real-world applications. In addition, we tested the compatibility of mytool, and the results showed that mytool could improve the performance of existing tools such as MOPT and QSYM.
In this paper, we propose a novel face alignment method that trains deep convolutional network from coarse to fine. It divides given landmarks into principal subset and elaborate subset. We firstly keep a large weight for principal subset to make our
network primarily predict their locations while slightly take elaborate subset into account. Next the weight of principal subset is gradually decreased until two subsets have equivalent weights. This process contributes to learn a good initial model and search the optimal model smoothly to avoid missing fairly good intermediate models in subsequent procedures. On the challenging COFW dataset [1], our method achieves 6.33% mean error with a reduction of 21.37% compared with the best previous result [2].
