Аннотация:Multitask learning is critical for the development of artificial general intelligence. Unfortunately, sequential learning of several tasks in artificial neural networks (ANNs) often results in catastrophic forgetting of previous knowledge. We propose a method of multitask learning based on the knowledge fusion of two ANNs. The method assumes pruning unessential weights in one ANN and replacing them with the weights of another network. This procedure makes it possible for the resulting network to have multitasked abilities. The proposed method does not require previous training data. The method is tested on convolutional spiking neural networks in image classification tasks. The experiments are performed on freely available datasets in the SpykeTorch simulation framework.