{"repo":"google-research/morph-net","free":true,"listed":false,"github":"https://github.com/google-research/morph-net","clone":"git clone https://github.com/google-research/morph-net.git","description":"Fast & Simple Resource-Constrained Learning of Deep Network Structure","language":"Python","stars":1038,"topics":["neural-architecture-search","automl","tensorflow","deep-learning","python","machine-learning"],"license":"Apache-2.0","category":"machine-learning","readme_excerpt":"MorphNet: Fast & Simple Resource-Constrained Learning of Deep Network Structure [TOC] New: FiGS: Fine-Grained Stochastic Architecture Search FiGS, is a probabilistic approach to channel regularization that we introduced in Fine-Grained Stochastic Architecture Search. It outperforms our previous regularizers and can be used as either a pruning algorithm or a full fledged Differentiable Architecture Search method. This is the recommended way to apply MorphNet. In the below documentation it is referred to as the LogisticSigmoid regularizer. What is MorphNet? MorphNet is a method for learning deep network structure during training. The key principle is continuous relaxation of the network-structure learning problem. In short, the MorphNet regularizer pushes the influence of filters down, and once they are small enough, the corresponding output channels are marked for removal from the network. Specifically, activation sparsity is induced by adding regularizers that target the consumption of specific resources such as FLOPs or model size. When the regularizer loss is added to the training loss and their sum is minimized via stochastic gradient descent or a similar optimizer, the learning problem becomes a constrained optimization of the structure of the network, under the constraint represented by the regularizer. The method was first introduced in our CVPR 2018, paper \"MorphNet: Fast & Simple Resource-Constrained Learning of Deep Network Structure\". A overview of the approach as w","default_branch":null,"files":null,"tree":[],"storefront":"/r/google-research","claimed":false,"request_supported":{"post":"https://gitbuyer.com/r/google-research/morph-net/request-supported","requests":0},"note":"indexed from public GitHub; nothing is for sale on this page. Clone it from GitHub. Paid listings live at /search."}