Non-Differentiable Supervised Learning with Evolution Strategies and Hybrid Methods

Abstract

In this work we show that Evolution Strategies (ES) are a viable method for learning non-differentiable parameters of large supervised models. ES are black-box optimization algorithms that estimate distributions of model parameters; however they have only been used for relatively small problems so far. We show that it is possible to scale ES to more complex tasks and models with millions of parameters. While using ES for differentiable parameters is computationally impractical (although possible), we show that a hybrid approach is practically feasible in the case where the model has both differentiable and non-differentiable parameters. In this approach we use standard gradient-based methods for learning differentiable weights, while using ES for learning non-differentiable parameters - in our case sparsity masks of the weights. This proposed method is surprisingly competitive, and when parallelized over multiple devices has only negligible training time overhead compared to training with gradient descent. Additionally, this method allows to train sparse models from the first training step, so they can be much larger than when using methods that require training dense models first. We present results and analysis of supervised feed-forward models (such as MNIST and CIFAR-10 classification), as well as recurrent models, such as SparseWaveRNN for text-to-speech.

References in corpus (15)

• Adam: A Method for Stochastic Optimization
• Categorical Reparameterization with Gumbel-Softmax
• The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables
• Evolution Strategies as a Scalable Alternative to Reinforcement Learning
• To prune, or not to prune: exploring the efficacy of pruning for model compression
• Don't Decay the Learning Rate, Increase the Batch Size
• The State of Sparsity in Deep Neural Networks
• Xception: Deep Learning with Depthwise Separable Convolutions
• Variational Dropout Sparsifies Deep Neural Networks
• Simple random search provides a competitive approach to reinforcement learning
• Deep Rewiring: Training very sparse deep networks
• Exploring Sparsity in Recurrent Neural Networks
• Faster gaze prediction with dense networks and Fisher pruning
• On the Relationship Between the OpenAI Evolution Strategy and Stochastic Gradient Descent
• TF-Replicator: Distributed Machine Learning for Researchers

Cited by in corpus (4)

• Improving Neural Network Training in Low Dimensional Random Bases
• MLE-guided parameter search for task loss minimization in neural sequence modeling
• Reinforcement Learning with Chromatic Networks for Compact Architecture Search
• ES-ENAS: Efficient Evolutionary Optimization for Large Hybrid Search Spaces