Skip to docs navigation

05.02 Pretrained Models

The models used in previous experiments with ml5.js and p5.js were pretrained models.

Pretrained Model Considerations

Whenever using a pre-trained AI model there are many points to consider. First, you should ask, “Who trained the model and why?” What is a state actor? A for-profit business? Doctors Without Borders? Just because a model was trained by a for-profit entity does not make it “bad” just as if a model was trained by a non-profit entity does not make it “good”. It is important to consider the incentives of the trainer of the model and what inherent biases they may have. While a for-profit entity may have good intentions releasing a pretrained model to help the public, there will always remain questions of profit motive.

It is also important to consider the data that was used to train the model. Was permission given to use the data? Does the data contain a representative and unbiased sample of information? Often the datasets for machine learning models can be found on the respective GitHub repositories of the model, but not always.

Pretrained Model Sources

Further Reading