P. Michael Furlong, Nicole Sandra-Yaffa Dumont, Rika Antonova, Jeff Orchard, Chris Eliasmith
Autonomous systems that learn and explore over long horizons face a problem. Standard methods scale poorly in the number of observations, $n$, precluding sustained operation on bounded hardware. We show that compositional, high-dimensional vector representations inspired by neural computation address these constraints. We use these representations to construct a Bayesian optimization (BO) algorithm that operates in complex spaces and reduces the time and memory requirements compared to state-of-the-art BO algorithms on diverse tasks. Whereas standard methods incur $\mathcal O(n^3)$ time and $\mathcal O(n^2)$ memory complexity, our approach holds both at $\mathcal O(d^2)$ in the embedding dimension, which remains constant over the algorithm's lifetime. Our algorithm reduces compute time by 60–200× without loss in accuracy. Implementation on neuromorphic hardware reduces energy consumption per sample by 30–188×. These efficiencies stem from converting sample selection into continuous optimization on a compact domain, implementable by gradient methods or neural dynamics, enabling long-term, resource-bound, autonomous exploration.