Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Parallelized Multi-Agent Bayesian Optimization in Lava

Shay Snyder, Derek Gobin, Victoria Clerico, Sumedh R. Risbud, Maryam Parsa

TL;DR

Lava Multi-Agent Optimization (LMAO) provides a scalable framework for distributed Bayesian optimization within the Lava neuromorphic software stack by coordinating multiple asynchronous evaluators with a central optimizer. It offers an abstract Solver interface and per-agent port infrastructure to enable random and grid searches across varying precision, demonstrating improved runtime and exploration on an Ackley benchmark, a fixed-precision transductive spiking GNN, and a QUBO satellite scheduling problem. The results show that multi-agent configurations yield speedups and better diversification under latency constraints, highlighting LMAO's potential for large-scale neuromorphic and hardware-aware optimization. This work delivers an open-source, generalizable tool for parallel parameter searches that can accelerate neural architecture design, robotics control, and other high-dimensional optimization tasks in neuromorphic settings.

Abstract

In parallel with the continuously increasing parameter space dimensionality, search and optimization algorithms should support distributed parameter evaluations to reduce cumulative runtime. Intel's neuromorphic optimization library, Lava-Optimization, was introduced as an abstract optimization system compatible with neuromorphic systems developed in the broader Lava software framework. In this work, we introduce Lava Multi-Agent Optimization (LMAO) with native support for distributed parameter evaluations communicating with a central Bayesian optimization system. LMAO provides an abstract framework for deploying distributed optimization and search algorithms within the Lava software framework. Moreover, LMAO introduces support for random and grid search along with process connections across multiple levels of mathematical precision. We evaluate the algorithmic performance of LMAO with a traditional non-convex optimization problem, a fixed-precision transductive spiking graph neural network for citation graph classification, and a neuromorphic satellite scheduling problem. Our results highlight LMAO's efficient scaling to multiple processes, reducing cumulative runtime and minimizing the likelihood of converging to local optima.

Parallelized Multi-Agent Bayesian Optimization in Lava

TL;DR

Lava Multi-Agent Optimization (LMAO) provides a scalable framework for distributed Bayesian optimization within the Lava neuromorphic software stack by coordinating multiple asynchronous evaluators with a central optimizer. It offers an abstract Solver interface and per-agent port infrastructure to enable random and grid searches across varying precision, demonstrating improved runtime and exploration on an Ackley benchmark, a fixed-precision transductive spiking GNN, and a QUBO satellite scheduling problem. The results show that multi-agent configurations yield speedups and better diversification under latency constraints, highlighting LMAO's potential for large-scale neuromorphic and hardware-aware optimization. This work delivers an open-source, generalizable tool for parallel parameter searches that can accelerate neural architecture design, robotics control, and other high-dimensional optimization tasks in neuromorphic settings.

Abstract

In parallel with the continuously increasing parameter space dimensionality, search and optimization algorithms should support distributed parameter evaluations to reduce cumulative runtime. Intel's neuromorphic optimization library, Lava-Optimization, was introduced as an abstract optimization system compatible with neuromorphic systems developed in the broader Lava software framework. In this work, we introduce Lava Multi-Agent Optimization (LMAO) with native support for distributed parameter evaluations communicating with a central Bayesian optimization system. LMAO provides an abstract framework for deploying distributed optimization and search algorithms within the Lava software framework. Moreover, LMAO introduces support for random and grid search along with process connections across multiple levels of mathematical precision. We evaluate the algorithmic performance of LMAO with a traditional non-convex optimization problem, a fixed-precision transductive spiking graph neural network for citation graph classification, and a neuromorphic satellite scheduling problem. Our results highlight LMAO's efficient scaling to multiple processes, reducing cumulative runtime and minimizing the likelihood of converging to local optima.
Paper Structure (8 sections, 2 figures, 2 tables, 2 algorithms)

This paper contains 8 sections, 2 figures, 2 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Architecture of LMAO within the Lava software framework
  3. Results & Discussion
  4. Traditional Non-Convex Optimization
  5. Transductive Spiking Graph Neural Networks
  6. Satellite Scheduling with Quadratic Unconstrained Binary Optimization
  7. Conclusion
  8. Acknowledgements

Figures (2)

  • Figure 1: Multiple independently operating agents processing different hyperparameters from the central search algorithm with LMAO.
  • Figure 2: (A) The runtime latency of LMAO using BO on the Ackley function ackley1987model with varying amounts of manually induced delay. (B) The accuracy convergence of single and multi-agent BO for citation graph classification snyder2024transductive. (C) Grid search execution times with satellite scheduling lava across different numbers of processes.