hyperparameterとは 意味・読み方・使い方

名詞

hyperparameter (plural hyperparameters)

1. (Bayesian statistics) A parameter of a prior (as distinguished from inferred parameters of the model for the underlying system under analysis).

   Synonym: superparameter

   Coordinate terms: metaparameter, parameter, subparameter

   this model has four hyperparameters

   • 2003, Gary Koop, Bayesian Econometrics, John Wiley & Sons Ltd., page 19:

     The exact interpretation of these hyperparameters becomes clearer once you have seen their role in the posterior and, hence, we defer a deeper discussion of prior elicitation until the next section.

2. (machine learning) A parameter whose value is set before the learning process begins.

   Coordinate terms: metaparameter, parameter, subparameter

   • 2016, Ian Goodfellow, Yoshua Bengio, Aaron Courville, chapter 11, in Deep Learning, MIT Press, →ISBN, page 420:

     The ideal learning algorithm just takes a dataset and outputs a function, without requiring hand tuning of hyperparameters.

   • 2024 May 4, Alex Hern, Dan Milmo, quoting James Betker, “Danger and opportunity for news industry as AI woos it for vital human-written copy”, in The Guardian‎, →ISSN:

     “Model behaviour is not determined by architecture, hyperparameters, or optimizer choices,” he said, referring to the technical difficulties of training a language model. “It’s determined by your dataset, nothing else. […] ”

派生語

• hyperparametric

参考

• hyperprior

Further reading

• hyperparameter on Wikipedia.Wikipedia
• hyperparameter (machine learning) on Wikipedia.Wikipedia