We study the problem of argument systems, where a computationally weak verifier outsources the execution of a computation to a powerful but untrusted prover, while being able to validate that the result was computed correctly through a proof generated by the prover. In addition, the zero-knowledge property guarantees that proof leaks no information about the potential secret input from the prover. Existing efficient zero-knowledge arguments with sublinear verification time require an expensive preprocessing phase that depends on a particular computation, and incur big overhead on the prover time and prover memory consumption.

This thesis proposes new constructions for zero-knowledge arguments that overcome the above problems. The new constructions require only a one time preprocessing and can be used to validate any computations later. They also reduce the overhead on the prover time and memory by orders of magnitude. We apply our new constructions to build a verifiable database system and verifiable RAM programs, leading to significant improvements over prior work.