Realistic Uncheatable Benchmarks

基于计算复杂性理论，设计现实不可欺骗基准的项目努力实现更复杂的计算机性能基准设计方法。目标是将随机化和密码学的思想与日常使用中现实的现有基准相结合。近年来，理论计算机科学取得了巨大的进步，特别是在随机化和相互作用方面。这些想法非常适合设计不可欺骗的基准。基准测试的目的是评估硬件和软件系统的性能和可靠性。总的来说，现有的基准设计主要集中在系统的“典型”程序和数据集上。使基准不受篡改的问题大多被忽视了。该项目正在进行的研究已经证明了将计算复杂性和密码学的思想应用于各种现有基准的可行性，并使它们更准确，更抗篡改，从而更值得信赖

The project on the design of realistic uncheatable benchmarks strives to achieve a more sophisticated design methodology of computer performance benchmarks, based on the theory of computational complexity. The goal is to integrate ideas from randomization and cryptography with realistic existing benchmark in everyday use. There has been tremendous progress in theoretical computer science in recent years, especially with regard to the power of randomization and interaction. These ideas are well suited to the design of uncheatable benchmarks. The purpose of benchmark testing is to estimate the performance and reliability of hardware and software systems. By and large existing benchmark design has focused on "typical" programs and data sets for the system. The problem of making benchmarks resistant to tampering has been mostly ignored. The on-going research in this project has demonstrated the feasibility of applying ideas from computational complexity and cryptography to a wide variety of existing benchmarks, and making them more accurate, resistant to tampering and thus more trustworthy.***