Delta-STN: Efficient Bilevel Optimization for Neural Networks using Structured Response Jacobians
arXiv:2010.13514
Abstract
Hyperparameter optimization of neural networks can be elegantly formulated as a bilevel optimization problem. While research on bilevel optimization of neural networks has been dominated by implicit differentiation and unrolling, hypernetworks such as Self-Tuning Networks (STNs) have recently gained traction due to their ability to amortize the optimization of the inner objective. In this paper, we diagnose several subtle pathologies in the training of STNs. Based on these observations, we propose the -STN, an improved hypernetwork architecture which stabilizes training and optimizes hyperparameters much more efficiently than STNs. The key idea is to focus on accurately approximating the best-response Jacobian rather than the full best-response function; we achieve this by reparameterizing the hypernetwork and linearizing the network around the current parameters. We demonstrate empirically that our -STN can tune regularization hyperparameters (e.g. weight decay, dropout, number of cutout holes) with higher accuracy, faster convergence, and improved stability compared to existing approaches.
arXiv commentPublished as a conference paper at Neurips 2020
References in corpus (22)
- Very Deep Convolutional Networks for Large-Scale Image Recognition
- Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
- Practical Bayesian Optimization of Machine Learning Algorithms
- Neural Architecture Search with Reinforcement Learning
- Improved Regularization of Convolutional Neural Networks with Cutout
- Tune: A Research Platform for Distributed Model Selection and Training
- Regularizing and Optimizing LSTM Language Models
- Scalable Bayesian Optimization Using Deep Neural Networks
- New insights and perspectives on the natural gradient method
- Flipout: Efficient Pseudo-Independent Weight Perturbations on Mini-Batches
- Hyperparameter optimization with approximate gradient
- Freeze-Thaw Bayesian Optimization
- Bilevel Programming for Hyperparameter Optimization and Meta-Learning
- Scalable Gradient-Based Tuning of Continuous Regularization Hyperparameters
- Stochastic Hyperparameter Optimization through Hypernetworks
- Learning Feature Hierarchies with Centered Deep Boltzmann Machines
- What is Local Optimality in Nonconvex-Nonconcave Minimax Optimization?
- Optimizing Millions of Hyperparameters by Implicit Differentiation
- Convergence of Learning Dynamics in Stackelberg Games
- Noisy Natural Gradient as Variational Inference
- On Solving Minimax Optimization Locally: A Follow-the-Ridge Approach
- Truncated Back-propagation for Bilevel Optimization