Parameter Prediction for Unseen Deep Architectures
This addresses computational inefficiency in training neural networks for researchers and practitioners, potentially leading to a new paradigm, though it is incremental in applying existing graph neural networks to a new task.
The paper tackles the problem of inefficient hand-designed parameter optimization in deep learning by using deep learning to predict parameters for unseen neural architectures, achieving 60% accuracy on CIFAR-10 with a ResNet-50 and top-5 accuracy approaching 50% on ImageNet.
Deep learning has been successful in automating the design of features in machine learning pipelines. However, the algorithms optimizing neural network parameters remain largely hand-designed and computationally inefficient. We study if we can use deep learning to directly predict these parameters by exploiting the past knowledge of training other networks. We introduce a large-scale dataset of diverse computational graphs of neural architectures - DeepNets-1M - and use it to explore parameter prediction on CIFAR-10 and ImageNet. By leveraging advances in graph neural networks, we propose a hypernetwork that can predict performant parameters in a single forward pass taking a fraction of a second, even on a CPU. The proposed model achieves surprisingly good performance on unseen and diverse networks. For example, it is able to predict all 24 million parameters of a ResNet-50 achieving a 60% accuracy on CIFAR-10. On ImageNet, top-5 accuracy of some of our networks approaches 50%. Our task along with the model and results can potentially lead to a new, more computationally efficient paradigm of training networks. Our model also learns a strong representation of neural architectures enabling their analysis.