Complex Image Classification With Micro Laser Neurons Integrated With DNN-Assisted Genetic Algorithm

The demand for more efficient computing systems is rapidly increasing with the advances in AI technologies. Spike-based machine learning promises significant computational efficiency while demanding minimal resources. We use micropillar lasers with integrated saturable absorber as the building blocks for a spiking neural network (SNN) due to their energy efficiency and ultra-fast computational capabilities [1]. The biomimetic properties of these microlasers have been shown to enable as ultrafast feature detection neurons for fully online, simple image classification tasks [2]. In this work, we take advantage of these biomimetic properties to demonstrate the computational capabilities of microlaser neurons (MLNs) for the ultrafast classification of the reduced MNIST dataset.