Recognition of Gurmukhi Handwritten City Names Using Deep Learning and Cloud Computing

Table 8

Comparison of SGD optimizer with three different batch sizes.

Batch size	Epochs	Training loss	Validation loss	Validation accuracy	Average validation accuracy
2	1	1.96	1.11	63.2	87.6%
	2	1.51	0.84	76.1
	3	1.25	0.83	75.3
	4	1.08	0.31	91.1
	5	1.03	0.30	90.7
	6	0.85	0.26	93.6
	7	0.88	0.21	95.0
	8	0.78	0.10	98
	9	0.68	0.10	97.2
	10	0.71	0.10	97.6
4	1	1.4	1.08	65.1	94.18%
	2	0.60	0.23	94.1
	3	0.38	0.15	95.7
	4	0.33	0.13	96
	5	0.27	0.11	97.3
	6	0.25	0.07	99.0
	7	0.22	0.07	99.0
	8	0.19	0.07	99.0
	9	0.18	0.07	98.8
	10	0.17	0.07	98.8
8	1	1.1	0.99	64.3	93.31%
	2	0.37	0.28	92.12
	3	0.24	0.15	95.3
	4	0.17	0.15	96.5
	5	0.13	0.13	96.3
	6	0.11	0.11	96.7
	7	0.10	0.09	97.0
	8	0.09	0.09	98.5
	9	0.08	0.08	98.2
	10	0.07	0.08	98.3