As far as I know, in order to train a machine learning (ML) model, both inputs and outputs of a train set should be known. Then, given a new input, the ML model tries to predict the output.
With TRNGs, only the output is known (there is no input). Can we still use ML models to predict the next TRNG output?
Asked
Active
Viewed 70 times
-1
Shannon
- 985
- 3
- 11
- 25
2 Answers
0
With TRNGs, only the output is known (there is no input). Can we still use ML models to predict the next TRNG output?
In order to predict, you first need to train an ML model. For training you need input data.
But for the usecase of TRNG(assuming true random number generator) check this article: link
taurz
- 194
- 2
- 9
0
True RNGs are inherently unpredictable, with any input (internal state) essentially unobservable. The output of a TRNG should be a Markov process: meaning the current output is completely unrelated to all past output.
You could consider past output as an 'input' in the sense it might give some clues about the internal state of the generator.
Then, you could use ML to attempt to find some link between current and past output, and if you do, one or more of the following might be true:
- your TRNG isn't truly random
- your ML has found a pattern where none exists. Like human brains, ML will fit patterns even where they don't exist. This is where validation datasets come in.
So basically, if you're confident of the TRNG, the answer is no.
virgesmith
- 762
- 1
- 7
- 18
-
Yes, I understand what you mean. Do you know what is the ML model that can be used for these kind of tasks (that uses part of output as input)? – Shannon Feb 19 '19 at 23:52
-
Check 'Time Series Forecasting'. That may be used in this problem. – Aditya Kansal Feb 20 '19 at 05:37