The ever-increasing demands of computationally expensive and high-dimensional problems require novel optimization methods to find near-optimal solutions in a reasonable amount of time. Bayesian Optimization (BO) stands as one of the best methodologies for learning the underlying relationships within multi-variate problems. This allows users to optimize time consuming and computationally expensive black-box functions in feasible time frames. Existing BO implementations use traditional von-Neumann architectures, in which data and memory are separate. In this work, we introduce Lava Bayesian Optimization (LavaBO) as a contribution to the open-source Lava Software Framework. LavaBO is the first step towards developing a BO system compatible with heterogeneous, fine-grained parallel, in-memory neuromorphic computing architectures (e.g., Intel's Loihi platform). We evaluate the algorithmic performance of the LavaBO system on multiple problems such as training state-of-the-art spiking neural network through back-propagation and evolutionary learning. Compared to traditional algorithms (such as grid and random search), we highlight the ability of LavaBO to explore the parameter search space with fewer expensive function evaluations, while discovering the optimal solutions.

翻译：日益增长的计算昂贵且高维问题需求催生了新型优化方法，以在合理时间内找到近似最优解。贝叶斯优化（Bayesian Optimization, BO）作为学习多变量问题内在关联的最佳方法论之一，使用户能够在可行时间范围内优化耗时且计算昂贵的黑盒函数。现有贝叶斯优化实现采用传统冯·诺依曼架构，其中数据与存储器相互分离。本研究将Lava贝叶斯优化（LavaBO）作为开源Lava软件框架的贡献引入，这是向开发兼容异构细粒度并行存内神经形态计算架构（如英特尔Loihi平台）的贝叶斯优化系统迈出的第一步。我们通过多项任务（如通过反向传播和进化学习训练最先进的脉冲神经网络）评估了LavaBO系统的算法性能。与传统算法（如网格搜索和随机搜索）相比，我们突显了LavaBO能以更少昂贵函数评估次数探索参数搜索空间，同时发现最优解的能力。