Author: ZmnSCPxj 2021-08-14 12:35:25
Published on: 2021-08-14T12:35:25+00:00
As Lightning usage continues to grow, it becomes increasingly important to delegate tasks to programs. One such task is setting channel fees for forwarding nodes. The proposed algorithm for this task is based on the theory that there is a single global maxima for price, where the earnings for a particular price point per unit time are highest. However, sampling actual data from the node channel fee is necessary since the topology of the network changes continuously, which interrupts any optimization algorithm used and brings in the possibility of noise.Hill Climbing is proposed as the algorithm to use, with evaluation of the function being a few days of sampling actual data from the node channel fee. Although it may not be the most efficient way to optimize, its anytime property makes it robust against topology changes. The proposal suggests using the weighted median of all the other peers of a peer as the initial starting point for the Hill Climbing algorithm.CLBOSS implements a variant of the above. In principle, a sapient being would still defeat any pre-sentient algorithm by taking the output of a pre-sentient algorithm, studying the entire problem, and making a single change that improves on the output of the pre-sentient algorithm.The article discusses the relationship between pre-sentient algorithms and general sapience. While a sentient being may discover that no improvement is possible, pre-sentient algorithms allow for more decisions to be left to policies. This frees up limited human willpower and lifespan to handle other decisions that pre-sentient algorithms cannot. The article uses the example of compilers to demonstrate how assemblers used to beat the best compilers before eventually looking at the compiler output to make small optimizations. However, this became less economically justifiable as compilers improved. Even if humanity were replaced by more rational beings, the same relationship between general sapience and pre-sentient algorithms should still hold. Therefore, moving decisions to pre-sentient algorithms can still be beneficial.
Updated on: 2023-06-01T18:47:13.294648+00:00