Reproduce “Differential Evolution and Bayesian Optimisation for Hyper-Parameter Selection in Mixed-Signal Neuromorphic Circuits Applied to UAV Obstacle Avoidance” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Llewyn Salt, David Howard, Giacomo Indiveri, Yulia Sandamirskaya
An optimized neuromorphic spiking neural network (LGMDNN) for UAV obstacle avoidance, utilizing event-based vision and incorporating biological adaptation and plasticity, achieved 100% accuracy on simple looming stimuli and enhanced robustness on real-world data. The study demonstrated Self-Adaptive Differential Evolution (SADE) as an effective method for tuning the network's numerous hyper-parameters.
Key insights
- Self-Adaptive Differential Evolution (SADE) consistently achieved superior fitness, accuracy, precision, and specificity for LGMDNN hyper-parameter tuning compared to other DE and Bayesian Optimization variants.
- The combination of membrane adaptation and Spike-Time Dependent Plasticity (STDP) substantially enhanced the LGMDNN's ability to detect looming objects, achieving perfect accuracy on various simple stimuli.
- The effectiveness of plasticity is highly context-dependent, with modest levels consistently beneficial, but excessive plasticity potentially hindering performance, especially in simulated environments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.