Elements of Learning
The key to generalization of learning from data, a.k.a. machine learning, are driven by the following three abstract characteristics
- Inductive biases favoring certain kinds of solutions
- Flexibility of modeling
- Data that is relevant to the targeted learning task
It is pretty interesting that this closely resembles the Fogg Behavior Model.
The one-to-one mapping of the model onto machine learning is
There is at least one difference I can think of. In the behavior model, motivation and ability compensate for each other. In machine learning, however, it is unclear whether inductive biases and flexibility can be traded for each other towards learning. If the model is not flexible enough, inductive biases may be futile.
Nevertheless, I think this abstract model needs two things:
- A better acronym.
- A way to drive actionable research agendas.