[Paper] Using an regression alignment technique to optimise coin Hash Difficulty
-
[b]Proposal for using an Artificially intelligent regression alignment technique to optimise the coin hash mining difficulty for fairness and smoothness. [/b]
This paper shows how it is possible to use artificial intelligence techniques to future predict the likely hash rate for the next set of blocks, by adjusting the “weightings” of a small number of historical temporal averages, and selecting the “weightings” that gives the best future prediction. The coin would use the future predicted hash rate value as part of its difficulty setting algorithm.
This means the coin would learn to automatically adjust the difficulty for peak mining periods, say overnight, or if pool switching re occurs. The optimisation technique is flexible ot varing circumstances, it would also tend to auto align to the correct block rate and align the smoothing to fairest to all miners “types” **.
It could report on how well it was doing, keeping a smooth and predictable block rate.
Unabated learning may have unforeseen effects, so it is envisaged the the AutoDifficulty alignment optimisation would be limited.
As previous hash and difficulty rate analysis showed, for Feathercoin during June, July for instance, a three temporal average covers the main period of miner / hash variation. It would be possible for the same alignment technique to be used to optimise the number and duration of the temporal periods.
Optimising Temporal period significance
( (A * T1) + (B * T2) + (C * T3) )/ ( (T1 + T2 + T3) * (A + B + C) ) = Combined average block time, used to calculate next difficulty.
where A, B, C are the adjustable weightings and T1, T2 & T3 are temporal block average time periods.
At each block the miner can change one variable a random small amount. It the change give a better future prediction over the last block, it is stored in the blockchain with the difficulty, for use in difficulty adjustment at the next block.
** Mining Types… i.e. Loyal miners, steady baseline who just mine one coin. Coin switchers, gamers using their PC at night to mine when not gaming, in the future ASICS farms, etc …
** apologises to any hippopotomonstrosesquippedaliophobia sufferers…
Ref:
Bob Willets
1998 - 2002
Eng.D. dans Maintenance, Condition Monitoring, Engineering - The University of ManchesterDevelopment of an Artificial Intelligence system for the classification of possible failure modes based upon several combined data sources, which included vibration and process data.
-
It’s a bit mind bending as I am adding the weightings on top of the combined average time per block being used for the difficulty calculation. The principle is correct, the equation will need jigging.
-
[quote name=“wrapper0feather” post=“46447” timestamp=“1387543611”]
It’s a bit mind bending as I am adding the weightings on top of the combined average time per block being used for the difficulty calculation. The principle is correct, the equation will need jigging.
[/quote]Seems a bit overkill. The problem is difficulty traps, not people deciding how to benefit from opportunities. I think we solved the difficulty trap. Also I disagree with the notion that the confirmation rate has to be perfectly the one in the design. That’s meaningless in an economy involving thousands to millions.
Also the more perfection tweaks we pile on, the harder it is for innovators to add new features because they have to make sure they either translate all the features correctly or that they don’t conflict with them.
