I am working on an algorithm which tries to improve existing predictive model. The predictive model is associated with several objectives (such as accuracy or model size) that can be optimized. Let's say that the algorithm is guaranteed to improve or not deteriorate model size while keeping the accuracy unchanged. However, it is not guaranteed to provide the optimal solution and it is not known how far from the optimal solution we are.
How should such algorithm be called?
- "heuristics"
- "heuristics with a guarantee"
- "optimization algorithm, which guarantees that the value of the objective function will improve or stay the same"
I am aware of Is machine learning an heuristic method?, but did not find the discussion particularly helpful.
