The Importance of Modular Structure in Artificial Intelligence Algorithm Evaluation

With the fast development of Artificial Intelligence (AI) algorithm, the AI algorithm evaluation approach, which only uses model performance as the evaluation criterion, is not appropriate anymore. Firstly, the performance-based evaluation strategy is difficult to deeply explore the fundamental contribution of the internal structure of AI algorithms to the problem solving; secondly, a complete process needs to be designed for each evaluation, which leads to a large amount of time and energy invested by researchers. In order to evaluate AI algorithms in a more efficient and standardized way, this paper combs through the current patterns of AI algorithm evaluation and deconstructs them into highly modular components to meet different experimental needs.