Controlled Random Search Improves Hyper-Parameter Optimization
Gowtham Muniraju, Bhavya Kailkhura, Jayaraman J. Thiagarajan, Peer‐Timo Bremer · arXiv (Cornell University) · 2018
Sampling one or more effective solutions from large search spaces is a recurring idea in computer vision, and sequential optimization has become the prevalent solution. Typical examples are hyper-parameter optimization in deep learning and sample mining in predictive modeling tasks. Existing solutions attempt to trade-off between global exploration and local exploitation, wherein the initial exploratory sample is critical to their success. While discrepancy-based samples have become the \textit{de facto} approach for exploration, results from computer graphics show that coverage-based designs, i.e. Poisson disk sampling (PDS), can be an effective alternative. However, in order to successfully employ PDS, originally developed for $2$-d image analysis, in computer vision problems, we propose fundamental advances, including defining a new parameterized PDS with maximal coverage, and developing algorithms for efficient sample synthesis. Using experiments in hyper-parameter optimization, we show that PDS samples significantly outperform existing exploratory sampling methods in both blind exploration, and sequential search with Bayesian optimization.