Plenary: Coding for Distributed Information Systems: Adversarial Sources and Approximated Decoding
Coding is known as an effective approach to deal with some of the major challenges in distributed information systems in terms of security of the network, privacy of the data, and efficiency of the resource management. The current solutions are often based on expanding the existing coding tools and techniques, still following the conventional mindset. In this talk, we argue that there are some major scenarios and thread models that do not conform with the traditional rules. For example: (1) Coding techniques are often designed for "exact error-free decoding". However, in some scenarios, e.g. straggler-resistance computing for machine learning, only an "approximate decoding" is enough, (2) In conventional coding techniques, "input sources" are supposed to be conveyed and decoded correctly. However, in some scenarios (e.g. sharded blockchains), some parts of "input symbols" are chosen and distributed adversarially to prevent "other input symbols" from being recovered correctly. In this talk, we will review some solutions and bounds for those scenarios and discuss some open problems.