The idea
Early experiments should sometimes maximize diversity of learning rather than incremental performance. Once a promising region appears, deeper optimization becomes more useful.
Worked example · illustrative
Finding exercise you’ll keep doing
Situation. You want to exercise more regularly. You’ve made a running plan and keep adjusting the distance and pace, but you rarely get out for a run.
The tempting conclusion. Maybe a better running plan would help me stick with it.
What changes the view. Running might not be the best fit for you right now. Trying a few walks, swims, or group classes could tell you more than another change to the plan. You might find that you enjoy a class with other people, or that a walk at lunchtime is easier to fit into your day.
What this supports. Once you find something you enjoy and can make time for, it may be worth building a routine around it. Then details like when to go or how much to do have a clearer purpose. You don’t have to keep trying everything, though a change in your schedule or interests might be a reason to try something else later.
Useful when
Watch for
Exploration can become endless experimentation. Concentrate effort when evidence identifies a promising direction, but reconsider it if the situation changes.
Where this came from
Applied machine learning, Research