On the Computational Efficiency of Training Neural Networks
It addresses the fundamental problem of computational efficiency in neural network training for researchers and practitioners, offering incremental improvements with theoretical and practical insights.
The paper revisits the computational complexity of training neural networks, providing both positive and negative results, including new provably efficient and practical algorithms for training certain types of neural networks.
It is well-known that neural networks are computationally hard to train. On the other hand, in practice, modern day neural networks are trained efficiently using SGD and a variety of tricks that include different activation functions (e.g. ReLU), over-specification (i.e., train networks which are larger than needed), and regularization. In this paper we revisit the computational complexity of training neural networks from a modern perspective. We provide both positive and negative results, some of them yield new provably efficient and practical algorithms for training certain types of neural networks.