LAUSR

Meta-learning can sum up the experience from the learned tasks, and solve the problems of datasets scarce or expensive and insufficient model generalization ability, which is the inadequate of traditional machine learning. Metalearning can obtain a well initial model, which can quickly generalize after a few adjustments to solve new tasks with good performance. However, model-agnostic metalearning (MAML) faces the problem of convergence difficulties, and easily prone to concussion late in training. We employ fractional order to MAML, which can retain past task gradient for a better stabilized and generalized model. We proposed a method based on fractional order and MAML (FracMAML), which can apply to a method based on metalearning, such as MAML and Reptile. Our proposed method can obtain a better initial model via fractional order. We take advantage of metalearning on few-shot problems to solve the aerospace targets classification problem, and proposed an Aerospace dataset. We employed FracMAML on MiniImagenet and Omniglot, which obtain state-of-the-art accuracy than some classical metalearning methods, such as MAML, Reptile, and so on. Finally, we verify the FracMAML method on aerospace targets classification task based on our Aerospace dataset, which performed well; further, verify the versatility of our algorithm.