Batched Gaussian Process Bandit Optimization via Determinantal Point Processes
About
Gaussian Process bandit optimization has emerged as a powerful tool for optimizing noisy black box functions. One example in machine learning is hyper-parameter optimization where each evaluation of the target function requires training a model which may involve days or even weeks of computation. Most methods for this so-called "Bayesian optimization" only allow sequential exploration of the parameter space. However, it is often desirable to propose batches or sets of parameter values to explore simultaneously, especially when there are large parallel processing facilities at our disposal. Batch methods require modeling the interaction between the different evaluations in the batch, which can be expensive in complex scenarios. In this paper, we propose a new approach for parallelizing Bayesian optimization by modeling the diversity of a batch via Determinantal point processes (DPPs) whose kernels are learned automatically. This allows us to generalize a previous result as well as prove better regret bounds based on DPP sampling. Our experiments on a variety of synthetic and real-world robotics and hyper-parameter optimization tasks indicate that our DPP-based methods, especially those based on DPP sampling, outperform state-of-the-art methods.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Quadratic Assignment Problem | QAPLIB nug instances | QAP Cost3.90e+3 | 13 | |
| Flowshop Scheduling Problem | FSP-car5 | Mean Objective Value7.80e+3 | 9 | |
| Quadratic Assignment Problem | QAP-chr12a | Mean Cost1.47e+4 | 9 | |
| Traveling Salesman Problem | TSP burma14 | Mean Tour Length3.79e+3 | 9 | |
| Flowshop Scheduling Problem | FSP-hel2 | Mean Performance142.5 | 8 | |
| Flowshop Scheduling Problem | FSP-reC19 | Mean Objective Value2.26e+3 | 8 | |
| Quadratic Assignment Problem | QAP-esc32a | Mean Cost276.5 | 8 | |
| Traveling Salesman Problem | TSP bayg29 | Mean Solution Value2.73e+3 | 8 | |
| Traveling Salesman Problem | TSP-att48 | Mean Tour Length3.95e+4 | 8 |