What is the way to protect against intruders in a group of distributed computing?

Question

Distributed Computing

Hello to you all.
Background: There is a large network of people who volunteer their computing power in exchange for a reward for each subproblem they solve. A common large task is loaded into the network, divided into subtasks, and these subtasks of low complexity are distributed among the nodes.
Problem: how to protect yourself from unscrupulous users who will solve problems incorrectly and send incorrect answers? We need a probability close to 100% that the wrong answer will not pass into the system.
What I just thought! What algorithm can solve this problem?
Personally, for now, I only have the option of loading an encrypted program into the user's RAM (so that its code cannot be changed at runtime), which would make sure that he does not really cheat and calculates the task. How feasible is this and is it possible to achieve close to 100% protection this way?
Tasks most often involve graphics processing. Users only use gpu, no cpu. Subtasks are simple and are usually calculated in 20-30 ms.

Answer 1

Calculate the same task for different people several times and check it
. It will also help to find errors in calculations.

Answer 2

Sign requests with your public key, which is contained in the program.
If the signature is invalid, the task is not completed.
Give the task to three clients and check that the result is the same for all three.
Any violation (invalid key or at least one mismatch) - remove the task from these three and give it to the other three without reporting anything to the client (that there were errors, etc.).

Answer 3

Why not just check if it's correct?