Delta-STN: Efficient Bilevel Optimization for Neural Networks using Structured Response Jacobians
This work addresses hyperparameter optimization for neural network practitioners, offering an incremental improvement over prior hypernetwork approaches.
The paper tackled the problem of hyperparameter optimization in neural networks by proposing Delta-STN, an improved hypernetwork architecture that stabilizes training and optimizes hyperparameters more efficiently than existing methods like Self-Tuning Networks, achieving higher accuracy, faster convergence, and improved stability in tuning regularization hyperparameters.
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 $Δ$-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 $Δ$-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.