Provably efficient neural network representation for image classification
arXiv:1711.04606
Abstract
The state-of-the-art approaches for image classification are based on neural networks. Mathematically, the task of classifying images is equivalent to finding the function that maps an image to the label it is associated with. To rigorously establish the success of neural network methods, we should first prove that the function has an efficient neural network representation, and then design provably efficient training algorithms to find such a representation. Here, we achieve the first goal based on a set of assumptions about the patterns in the images. The validity of these assumptions is very intuitive in many image classification problems, including but not limited to, recognizing handwritten digits.
References in corpus (10)
- Efficient classical simulation of slightly entangled quantum computations
- Entanglement renormalization
- Tensorizing Neural Networks
- Classical simulation of quantum many-body systems with a tree tensor network
- Algorithms for entanglement renormalization
- Unsupervised Generative Modeling Using Matrix Product States
- Equivalence of restricted Boltzmann machines and tensor network states
- Deep Learning and Quantum Entanglement: Fundamental Connections with Implications to Network Design
- Analysis and Design of Convolutional Networks via Hierarchical Tensor Decompositions
- Entanglement Entropy of Target Functions for Image Classification and Convolutional Neural Network