TABLE 2.

State-of-the-art accuracy obtained with the STDP learning rule is tabulated for different numbers of epochs and output neurons.

This work			Past studies
Neurons	Epochs	Accuracy (%) (μ±σ)	Neurons	Epochs	Accuracy (%)	References
10	1	61.4 ± 0.78	10	1	60	Querlioz et al., 2013
50	1	78.84 ± 1.28	50	1	76.8	Guo et al., 2019
50	3	81.3 ± 1.76	50	3	77.2	Boybat et al., 2018
			50	1	78.55	Demin et al., 2021
			50	3	81	Querlioz et al., 2013
			50	-	83.03	Oh et al., 2019
100	3	84.74 ± 1.08	100	3	82.9	Diehl and Cook, 2015
			100	1	89.15	Demin et al., 2021
			200	17	91.63	Oh et al., 2019
300	3	89.08 ± 0.49	300	3	93.5	Querlioz et al., 2013
400	3	89.26 ± 0.54	400	3	87	Diehl and Cook, 2015
500	5	90.56 ± 0.27

The performance achieved by training SNN with the VDSP rule is tabulated for various network sizes (number of output neurons) and epochs. Each experiment was repeated with five different initial conditions, and the accuracies are reported as (Mean ± S.D.). Compared with the hardware-independent approach of pair based STDP (Diehl and Cook, 2015), we achieved 84.74 ± 1.08% for a network of 100 output neurons trained over three epochs. For a network of 400 output neurons trained over three epochs, we achieved 89.26 ± 0.54%.