Scope and Aim:

Machine learning is inherently a multi-objective task. A machine learning method usually consists of selecting a candidate model and tuning the parameters of the model using the learning algorithm and available data. Normally, there are multiple objectives to be considered in model selection, such as approximation error, variance, model complexity, and fairness, which are likely to conflict with each other, e.g., reducing the approximation error often leads to an increase in model complexity. Traditional methods either consider only one of the objectives or aggregate multiple objectives into a single-objective fitness function, which results in only one solution and therefore little knowledge of the task can be gained, and it is also not easy to set a suitable coefficient to aggregate multiple objectives. Optimization is at the heart of many machine learning techniques. With the boom of the research on evolutionary multi-objective optimization, multi-objective machine learning has gained new impetus. Multi-objective evolutionary algorithms can provide a set of trade-off solutions to meet the multiple criteria for optimizing machine learning models. Decision-makers can make a better choice from these trade-offs through extracting knowledge about the task and incorporating their preferences.

This special session aims to bring together theories and applications of evolutionary multi-objective techniques to machine learning tasks. In this sense, this special session aims to be a meeting place for researchers in the field of evolutionary multi-objective machine learning, to enrich both disciplines by means of the hybridization of state-of-the-art approaches from those domains.