Event-Driven Continuous STDP Learning With Deep Structure for Visual Pattern Recognition

Abstract

Human beings can achieve reliable and fast visual pattern recognition with limited time and learning samples. Underlying this capability, ventral stream plays an important role in object representation and form recognition. Modeling the ventral steam may shed light on further understanding the visual brain in humans and building artificial vision systems for pattern recognition. The current methods to model the mechanism of ventral stream are far from exhibiting fast, continuous, and event-driven learning like the human brain. To create a visual system similar to ventral stream in human with fast learning capability, in this paper, we propose a new spiking neural system with an event-driven continuous spike timing dependent plasticity (STDP) learning method using specific spiking timing sequences. Two novel continuous input mechanisms have been used to obtain the continuous input spiking pattern sequence. With the event-driven STDP learning rule, the proposed learning procedure will be activated if the neuron receive one pre- or post-synaptic spike event. The experimental results on MNIST database show that the proposed method outperforms all other methods in fast learning scenarios and most of the current models in exhaustive learning experiments.

Biologically plausible deep learning — But how far can we go with shallow networks?

“…Examples include the use of convolutional layers [28,13,29,30] (and tables therein), dendritic computations [31,32,12] or backpropagation approximations such as feedback alignment [11,33,34,35,36,14] equilibrium propagation [37], membrane potential based backpropagation [38], restricted Boltzmann machines and deep belief networks [39,40], (localized) difference target propagation [41,14], using reinforcement-signals [42,43] or approaches using predictive coding [44]. Many models implement spiking neurons to stress bio-plausibility [45,46,47,48,49,13] (and tables therein) or coding efficiency [50]. The conversion of DNNs to spiking neural networks (SNN) after training with backpropagation [51] is a common technique to evade the difficulties of training with spikes.…”

Memristive Spiking Neural Networks Trained with Unsupervised STDP

“…Lateral inhibition and adaptive threshold voltages are required in ref. [22] as well. Other SNNs have similar requirements.…”

“…In ref. [22], the spike time of an input neuron is inversely proportional to the input signal. In this paper, their relationship is described as follows:…”

“…Note that the encoding schemes of above STDP-based SNNs are all rate-based, i.e., the information is transmitted by spike rate. Whereas, some researchers have showed that time-based SNNs may be more efficient in processing and hardware implementation [21,22]. Moreover, they think the first spike is enough to transmit information.…”

“…Kheradpisheh et al presented a time-based deep convolutional SNN and gave corresponding unsupervised learning rules [12]. Liu et al proposed a time-based hierarchical structure based on supervised STDP learning rules [22].…”