A Neuromorphic Model for Image Recognition using SNN

Distinctive properties of biological neural networks and recent advances in bottom-up approaches toward a better biologically plausible neural network

“…For certain tasks, such as predicting the digit annotation from the images drawn from the MNIST dataset (LeCun et al, 2010 ) after supervised learning, the ANN can achieve the best precision, while the SNN may not be able to outperform it. However, when implemented in hardware, SNNs have a considerably greater advantage in terms of power consumption, as observed in modern neuromorphic hardware (Cao et al, 2015 ; Pfeiffer and Pfeil, 2018 ; Cui et al, 2019 ; Kornijcuk et al, 2019 ; Kabilan and Muthukumaran, 2021 ; Parker et al, 2022 ). In addition, as mentioned in Section 3.1, SNNs may have the advantage of dealing with intermittently activated inputs (Pfeiffer and Pfeil, 2018 ).…”

“…Such a strategy is advantageous for richer dynamics and encoding capacity as well as lower power consumption by considering silence (off period) as another piece of information (Cao et al, 2015 ; Pfeiffer and Pfeil, 2018 ). Therefore, spiking neural networks (SNN) has become an essential type of ANN and are widely utilized in neuromorphic engineering (Kornijcuk et al, 2019 ; Kabilan and Muthukumaran, 2021 ; Parker et al, 2022 ). Because various models can describe a neuron's spike activity and each spike can represent distinctive information depending on the coding scheme, we can expect a much larger diversity of neuronal activation processes compared to ANN.…”

Organic All-Photonic Artificial Synapses Enabled by Anti-Stokes Photoluminescence

“…9−11 For example, in the field of image or video recognition, neuromorphic models using convolution and spike time-dependent plasticity (STDP) can provide a fast learning rate and low power consumption. 12 Organic artificial synapses are promising technologies for future neuromorphic electronics due to their advantages of easily tunable physicochemical properties, high compatibility with solution processes, low energy consumption, and mechanical flexibility/stretchability with a low elastic modulus. 13−19 However, electrical synapses inevitably require extra electrical consumption, although many efforts have been made to reduce their driving voltages.…”

“…Artificial intelligence (AI) is the science and engineering to develop intelligent machines for simulating and expanding human intelligence. − Nowadays, the traditional Von Neumann computer is in the dilemma of limited operating speed and large energy consumption ( E ) due to the separated processor and memory. , In comparison, human brains can process and memorize information simultaneously through the manipulation of synaptic weight. − Thus, the simulation of synaptic behaviors for the artificial neural network may be a solution to the bottleneck of Von Neumann, allowing more efficient and parallel processing. − For example, in the field of image or video recognition, neuromorphic models using convolution and spike time-dependent plasticity (STDP) can provide a fast learning rate and low power consumption …”