Fig. 4: High-order STDP learning rules-based spiking neural networks (SNNs) for image classification tasks. | Nature Communications

Fig. 4: High-order STDP learning rules-based spiking neural networks (SNNs) for image classification tasks.

From: Bioinspired high-order in-sensor spatiotemporal enhancement in van der Waals optoelectronic neuromorphic electronics

Fig. 4

a Schematic of the high-order STDP-enabled SNN based on the proposed vdW phototransistors with Triplet-STDP behavior. The right panel shows the image classification workflow with the alphanumeric dataset. b Training accuracy of image classification as a function of iteration based on Triple-STDP and the corresponding effects of different fine-tuning factor coefficients. The inset shows a local zoomed-in accuracy plot, where Triplet-STDP exhibits the highest training accuracy when the fine-tuning factor is 6%. c Comparison of training accuracy between Paired-STDP and Triplet-STDP under varying fine-tuning factors, indicating that the generalization performance of the SNN can be improved by the tuning factor within a certain range. d Training accuracy comparison between Paired-STDP and Triplet-STDP learning rules for classifying characters with visually similar features. For example, as to the classification of confusing items of digit 1 and letter I, the feature extraction capability of Triplet-STDP SNN can be distinctly enhanced without complicating the SNN model.

Back to article page