EARL: Energy-Aware Optimization of Liquid State Machines for Pervasive AI

TL;DR

The paper addresses the challenge of deploying Liquid State Machines on energy-constrained devices by explicitly optimizing for both accuracy and energy. It introduces EARL, a hybrid framework that combines Bayesian optimization with reinforcement learning to guide candidate selection and includes an adaptive early termination mechanism to reduce evaluations. EARL employs Sobol initialization, Gaussian Process surrogate modeling, and Expected Improvement to explore the hyperparameter landscape, while an RL agent selects among BO-proposed configurations using a reward $r(x) = f_1(x) - \alpha f_2(x)$ to balance performance and power. Experiments on three datasets show that EARL achieves 6–15% higher accuracy, 60–80% lower energy per sample, and up to an order of magnitude faster optimization than baselines like Optuna and Ray Tune, demonstrating the practicality of energy-aware adaptive search for on-device LSMs.

Abstract

Pervasive AI increasingly depends on on-device learning systems that deliver low-latency and energy-efficient computation under strict resource constraints. Liquid State Machines (LSMs) offer a promising approach for low-power temporal processing in pervasive and neuromorphic systems, but their deployment remains challenging due to high hyperparameter sensitivity and the computational cost of traditional optimization methods that ignore energy constraints. This work presents EARL, an energy-aware reinforcement learning framework that integrates Bayesian optimization with an adaptive reinforcement learning based selection policy to jointly optimize accuracy and energy consumption. EARL employs surrogate modeling for global exploration, reinforcement learning for dynamic candidate prioritization, and an early termination mechanism to eliminate redundant evaluations, substantially reducing computational overhead. Experiments on three benchmark datasets demonstrate that EARL achieves 6 to 15 percent higher accuracy, 60 to 80 percent lower energy consumption, and up to an order of magnitude reduction in optimization time compared to leading hyperparameter tuning frameworks. These results highlight the effectiveness of energy-aware adaptive search in improving the efficiency and scalability of LSMs for resource-constrained on-device AI applications.

Figures (3)

• Figure 1: Classification accuracy across all optimization models
• Figure 2: Energy consumption across all optimization models.
• Figure 3: Pareto Front vs All Trials.