Table 2.

Comparison of accuracy of different models on PI MNIST.

Network	# units in HLs	Test accuracy (%)
ANN (Srivastava et al., 2014)	800	98.4
ANN (Srivastava et al., 2014), Drop-out	4096–4096	98.99
ANN (Wan et al., 2013), Drop-connect	800–800	98.8
ANN (Goodfellow et al., 2013), maxout	240 × 5–240 × 5	99.06
SNN (O'Connor et al., 2013)a,b	500–500	94.09
SNN (Hunsberger and Eliasmith, 2015)a	500–300	98.6
SNN (Diehl et al., 2015)	1200–1200	98.64
SNN (O'Connor and Welling, 2016)	200–200	97.8
SNN (SGD, This work)	800	[98.56, 98.64, 98.71]*
SNN (SGD, This work)	500–500	[98.63, 98.70, 98.76]*
SNN (ADAM, This work)	300–300	[98.71, 98.77, 98.88]*

We compare only to models that do not use unsupervised pre-training or data augmentation, with the exception of O'Connor et al. (2013) and Hunsberger and Eliasmith (2015).

^apretraining,

^bdata augmentation,

^*[min, average, max] values over epochs [181, 200].