AI Dictionary › Fondamenti AI
Iperparametro
A hyperparameter is a configuration value of a model or its training process that is chosen before training begins and is not automatically modified by the learning algorithm, unlike actual parameters, such as weights, which are progressively updated on the data. Typical examples include the number of layers, the size of hidden layers, the learning rate, the batch size and the number of epochs.
The choice of hyperparameters directly affects both the final quality of the model and the computational cost of training: values that are too aggressive can make training unstable, values that are too conservative can make learning extremely slow or insufficient. Finding the best combination often requires training multiple variants of the same model and comparing results, a process called hyperparameter search.
It is a practice present at every stage of developing artificial intelligence models, from small classifiers to large language models, where decisions such as the initial learning rate or the schedule for reducing it can determine whether a multimillion-dollar training run converges correctly or fails.
The prefix "hyper" indicates that these values sit at a level above the model's own parameters: while parameters are learned from data, hyperparameters govern how that learning happens, which is why they are decided in advance by whoever designs or trains the model.
From our network
AGORÀ Intelligence: Enterprise AI Governance Platform
Govern AI at scale: policies, adoption and measurable results on your data. Built for boards and C-suite.
Visit agora-intelligence.com →From the Agora Intelligence blog
📱 Download the Android app (beta) iOS coming soon
Say what you mean. Get what you need.
Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.