Table 2.

Parameters used for the continuous-time spiking neural network simulation implementing eRBP.

Nd	Number of data neurons	All networks	784
Nh	Number of hidden neurons	All networks	100,200,400,1000
Nl	Number of label neurons	All networks	10
NE+	Number of positive error neurons	All networks	10
NE−	Number of negative error neurons	All networks	10
Np	Number of prediction neurons	All networks	10
σ	Poisson noise weight	eRBP+	50· 10−3 nA
		eRBP×	0· 10−3 nA
p	Blank-out probability	eRBP+	1.0
		eRBP×	0.45
τrefr	Refractory period	Prediction and hidden neurons	3.9 ms
		Data neurons	4.0 ms
τsyn	Synaptic Time Constant	All synapses	4 ms
gV	Leak conductance state V	Prediction and hidden neurons	1 nS
gU	Leak conductance state U	Prediction and hidden neurons	5 nS
C	Membrane capacitance	All neurons	1 pF
VT	Firing threshold	Prediction and Hidden neurons	100 mV
$V_T^E$		Error neurons	100 mV
Ntrain	Number of training samples used	All figures	50000
Ntest	Number of training samples used	Table 1 eRBP+, eRBP×	10000
		Table 2 eRBP+, eRBP×	1000
		Table 2 RBP, BP	10000
Ttrain	Training sample duration	All models	100 mV
Ttest	Testing sample duration	Table 1, Figure 4	500 ms
		Table 2	250 ms
wh, wd, wp, g	Initial weight matrix	RBP, BP	$U(\sqrt{\frac{6}{\#rows+\#cols}})\text{nA}$
		eRBP+	$U(\sqrt{\frac{6}{\#rows+\#cols}})\text{nA}$
		eRBP×	$U(\sqrt{\frac{7}{\#rows+\#cols}})\text{nA}$
wE		eRBP+, eRBP×	90· 10−3nA
wL+		eRBP+, eRBP×	90· 10−3nA
wL−		eRBP+, eRBP×	−90· 10−3nA
bmin,bmax		eRBP+, eRBP×	−1.15, 1.15 nA
	2nd hidden layer	eRBP+, eRBP×	-25, 25 nA
	Figure 6	eRBP+, eRBP×	−0.6, 0.6 nA
β	Data neuron input scale	eRBP+, eRBP×	0.5
γ	Data neuron input threshold	eRBP+, eRBP×	−0.215
η	Learning Rate	eRBP+	6· 10−4nS
		eRBP×	10· 10−4nS
		RBP, BP	0.4/nbatch
nbatch	Minibatch size	RBP(100), BP(100)	100
		RBP(1), BP(1)	1