Multi-task learning employs shared representation of knowledge for learningmultiple instances from the same or related problems. Time series predictionconsists of several instances that are defined by the way they are broken downinto fixed windows known as embedding dimension. Finding the optimal values forembedding dimension is a computationally intensive task. Therefore, weintroduce a new category of problem called dynamic time series prediction thatrequires a trained model to give prediction when presented with differentvalues of the embedding dimension. This can be seen a new class of time seriesprediction where dynamic prediction is needed. In this paper, we propose aco-evolutionary multi-task learning method that provides a synergy betweenmulti-task learning and coevolution. This enables neural networks to retainmodularity during training for building blocks of knowledge for differentinstances of the problem. The effectiveness of the proposed method isdemonstrated using one-step-ahead chaotic time series problems. The resultsshow that the proposed method can effectively be used for different instancesof the related time series problems while providing improved generalisationperformance.
展开▼