Combining Spiking Neural Networks with Artificial Neural Networks for Enhanced Image Classification

With the continued innovation of deep neural networks, spiking neural networks (SNNs) that more closely resemble biological brain synapses have attracted attention because of their low power con-sumption. Unlike artificial neural networks (ANNs), for continuous data values, they must employ an encoding process to convert the values to spike trains, suppressing the SNN's performance. To avoid this degrada-tion, the incoming analog signal must be regulated prior to the encoding process, which is also realized in living things e.g., the basement mem-branes of humans mechanically perform the Fourier transform. To this end, we combine an ANN and an SNN to build ANN-to-SNN hybrid neu-ral networks (HNNs) that improve the concerned performance. To qualify this performance and robustness, MNIST and CIFAR-10 image datasets are used for various classification tasks in which the training and encod-ing methods changes. In addition, we present simultaneous and separate training methods for the artificial and spiking layers, considering the en-coding methods of each. We find that increasing the number of artificial layers at the expense of spiking layers improves the HNN performance. For straightforward datasets such as MNIST, similar performances as ANN's are achieved by using duplicate coding and separate learning. However, for more complex tasks, the use of Gaussian coding and simultaneous learning is found to improve the accuracy of the HNN while lower power consump-tion.