Multi-layered Spiking Neural Network with Target Timestamp Threshold Adaptation and STDP

Abstract

Spiking neural networks (SNNs) are good candidates to produce ultra-energy-efficient hardware. However, the performance of these models is currently behind traditional methods. Introducing multi-layered SNNs is a promising way to reduce this gap. We propose in this paper a new threshold adaptation system which uses a timestamp objective at which neurons should fire. We show that our method leads to state-of-the-art classification rates on the MNIST dataset (98.60%) and the Faces/Motorbikes dataset (99.46%) with an unsupervised SNN followed by a linear SVM. We also investigate the sparsity level of the network by testing different inhibition policies and STDP rules.

SpikeSEG: Spiking Segmentation via STDP Saliency Mapping

“…The SpikeSEG network performs well on the synthetic dataset with accuracy values of 97% and mIoU of 74%, with a the results of all experiments shown in Table I . This accuracy is inline with results from [18], [23] with a slight drop in accuracy expected when coverting to a fully convolutional network. The mIoU accuracy provides a good return for (a) Input segmentation extremity based bounding box estimation, with mIoUs of >50% seen as acceptable and over 75% being close to state of the art in some application.…”

“…However, this method is still burdened with the training computational overhead and does little to utilise the efficiency of event driven computations. The SNN's Spike Time Dependent Plasticity (STDP) and spike-based back-propagation learning have been demonstrated to capture hierarchical features in SpikeCNNs [18]- [23] Both of these methods better equip the network to deal with event driven sensors, where the significant gains over CNNs could be realised.…”

“…It was also shown in [28] the importance of thresholds when converting deep neural networks to spiking, as the hierarchical layers clearly have a cascading affect. Research by Falez et al [23] showed using Time target threshold adaptation, within a STDP trained 2 Conv Layer network provided promising results. A method where an ideal time to fire is learned and used to adapt the thresholds according to try and achieve this.…”

“…STDP has been shown to be a useful learning mechanism for unsupervised learning with a biological context with good results in [18], [23] showing its ability within the 2-3 conv layer range to be able to learn extract useful features for classification type task. STDP as the name suggests, directly accounts for time explicitly, meaning it is well suited for learning from asynchronous sensor, with single layer networks having shown impressive results [22], [33], [34].…”

Bio-inspired digit recognition using reward-modulated spike-timing-dependent plasticity in deep convolutional networks

“…SNNs with single spike per neuron have got more popularity recently. Apart from the energy-efficiency, they have shown to be competitive if the parameter values are tuned well for the target task [59,60].…”