Day 8: Transfer learning - Impact of the number of hidden layers 4

Yesterday, I found out that additional hidden layer does not have any impact on the time to process each batch. Today, I want to find out whether it has any impact on the test loss and accuracy.

The result for one hidden layer,
epochs steps processing_time/batch training_loss test_loss accuracy

1/3 5 9.998 2.583 1.164 0.504
1/3 10 10.004 0.817 0.242 0.911
1/3 15 10.019 0.247 0.103 0.970
1/3 20 10.009 0.177 0.132 0.953
1/3 25 9.999 0.197 0.075 0.975
1/3 30 9.986 0.221 0.066 0.978
1/3 35 10.028 0.214 0.062 0.978
1/3 40 10.017 0.164 0.058 0.977
1/3 45 10.005 0.112 0.069 0.972
1/3 50 9.959 0.228 0.055 0.978
2/3 5 9.866 0.197 0.060 0.978
2/3 10 10.019 0.178 0.084 0.967
2/3 15 9.690 0.252 0.064 0.977
2/3 20 10.035 0.295 0.054 0.983
2/3 25 9.698 0.148 0.061 0.981
2/3 30 10.009 0.140 0.113 0.958
2/3 35 9.697 0.215 0.196 0.927
2/3 40 10.030 0.275 0.053 0.978
2/3 45 9.724 0.213 0.042 0.984
2/3 50 10.020 0.148 0.053 0.983
3/3 5 10.193 0.133 0.045 0.984
3/3 10 10.004 0.164 0.041 0.984
3/3 15 9.994 0.100 0.042 0.984
3/3 20 10.018 0.148 0.039 0.985
3/3 25 10.017 0.115 0.048 0.983
3/3 30 10.007 0.128 0.070 0.975
3/3 35 9.995 0.235 0.050 0.981
3/3 40 10.033 0.193 0.069 0.973
3/3 45 9.989 0.181 0.091 0.966
3/3 50 10.020 0.178 0.086 0.964

The result for two hidden layers,
epochs steps processing_time/batch training_loss test_loss accuracy

1/3 5 9.808 3.117 0.860 0.494
1/3 10 9.779 0.765 0.627 0.506
1/3 15 9.803 0.632 0.417 0.950
1/3 20 9.798 0.376 0.147 0.977
1/3 25 9.812 0.279 0.077 0.976
1/3 30 9.799 0.251 0.128 0.953
1/3 35 9.804 0.558 0.185 0.925
1/3 40 9.812 0.292 0.146 0.955
1/3 45 9.791 0.333 0.245 0.899
1/3 50 9.793 0.307 0.114 0.978
2/3 5 9.992 0.296 0.063 0.981
2/3 10 9.792 0.203 0.053 0.981
2/3 15 9.801 0.208 0.047 0.983
2/3 20 9.807 0.169 0.132 0.953
2/3 25 9.791 0.237 0.075 0.974
2/3 30 9.826 0.172 0.071 0.977
2/3 35 9.817 0.143 0.055 0.982
2/3 40 9.848 0.104 0.060 0.977
2/3 45 9.834 0.224 0.087 0.971
2/3 50 9.865 0.461 0.105 0.959
3/3 5 10.053 0.318 0.081 0.980
3/3 10 9.915 0.173 0.067 0.980
3/3 15 9.899 0.198 0.069 0.975
3/3 20 9.917 0.212 0.056 0.981
3/3 25 9.921 0.197 0.058 0.978
3/3 30 9.867 0.135 0.110 0.959
3/3 35 9.905 0.305 0.077 0.975
3/3 40 9.911 0.171 0.068 0.979
3/3 45 9.900 0.254 0.089 0.969
3/3 50 9.901 0.173 0.055 0.981

The result for three hidden layers,
epochs steps processing_time/batch training_loss test_loss accuracy

1/3 5 9.849 1.203 0.726 0.494
1/3 10 9.892 0.675 0.690 0.494
1/3 15 9.926 0.683 0.393 0.976
1/3 20 9.906 0.381 0.135 0.957
1/3 25 9.914 0.235 0.061 0.979
1/3 30 9.924 0.296 0.060 0.977
1/3 35 9.928 0.304 0.057 0.983
1/3 40 9.918 0.434 0.085 0.982
1/3 45 9.910 0.319 0.158 0.966
1/3 50 9.911 0.167 0.077 0.981
2/3 5 9.896 0.199 0.046 0.984
2/3 10 9.729 0.199 0.082 0.968
2/3 15 9.717 0.199 0.060 0.979
2/3 20 9.727 0.203 0.056 0.983
2/3 25 9.739 0.111 0.091 0.967
2/3 30 9.720 0.207 0.067 0.974
2/3 35 9.764 0.170 0.056 0.980
2/3 40 9.728 0.190 0.060 0.979
2/3 45 9.743 0.183 0.067 0.971
2/3 50 9.726 0.196 0.045 0.982
3/3 5 10.085 0.290 0.051 0.984
3/3 10 9.902 0.136 0.054 0.980
3/3 15 9.903 0.170 0.060 0.978
3/3 20 9.886 0.182 0.041 0.984
3/3 25 9.906 0.219 0.052 0.983
3/3 30 9.919 0.139 0.066 0.969
3/3 35 9.934 0.257 0.072 0.981
3/3 40 9.908 0.171 0.044 0.984
3/3 45 9.924 0.129 0.039 0.986
3/3 50 9.887 0.209 0.057 0.979

The result for five hidden layers,
epochs steps processing_time/batch training_loss test_loss accuracy

1/3 5 9.616 0.699 0.837 0.506
1/3 10 9.980 0.705 0.578 0.494
1/3 15 9.652 0.479 0.182 0.974
1/3 20 9.643 0.310 0.227 0.947
1/3 25 9.667 0.776 0.203 0.913
1/3 30 10.050 0.316 0.273 0.958
1/3 35 9.655 0.337 0.183 0.974
1/3 40 9.642 0.219 0.123 0.951
1/3 45 9.639 0.246 0.053 0.977
1/3 50 10.026 0.137 0.060 0.977
2/3 5 9.796 0.231 0.055 0.978
2/3 10 10.014 0.245 0.061 0.979
2/3 15 9.662 0.155 0.092 0.965
2/3 20 9.664 0.173 0.054 0.978
2/3 25 9.639 0.143 0.082 0.973
2/3 30 10.013 0.249 0.106 0.966
2/3 35 9.635 0.154 0.077 0.976
2/3 40 9.648 0.220 0.075 0.970
2/3 45 9.656 0.175 0.074 0.970
2/3 50 10.026 0.385 0.113 0.961
3/3 5 9.831 0.213 0.094 0.970
3/3 10 10.008 0.172 0.054 0.981
3/3 15 9.638 0.334 0.085 0.964
3/3 20 9.660 0.222 0.099 0.977
3/3 25 9.652 0.258 0.054 0.981
3/3 30 10.033 0.149 0.044 0.982
3/3 35 9.658 0.155 0.060 0.976
3/3 40 9.620 0.146 0.088 0.970
3/3 45 9.658 0.251 0.066 0.981
3/3 50 10.037 0.175 0.077 0.977

The result for ten hidden layers,
epochs steps processing_time/batch training_loss test_loss accuracy

1/3 5 9.891 0.728 0.694 0.494
1/3 10 9.858 0.697 0.686 0.494
1/3 15 9.832 0.785 0.693 0.494
1/3 20 9.852 0.693 0.676 0.506
1/3 25 9.866 0.676 0.540 0.506
1/3 30 9.856 0.650 0.634 0.845
1/3 35 9.859 0.556 0.309 0.947
1/3 40 9.848 0.301 0.106 0.955
1/3 45 9.861 0.199 0.099 0.961
1/3 50 9.853 0.585 0.147 0.921
2/3 5 9.858 0.558 0.360 0.956
2/3 10 9.676 0.780 0.065 0.978
2/3 15 9.719 0.385 0.316 0.943
2/3 20 9.685 0.516 0.491 0.693
2/3 25 9.689 0.566 0.286 0.960
2/3 30 9.705 0.319 0.074 0.981
2/3 35 9.689 0.311 0.154 0.962
2/3 40 9.696 0.340 0.136 0.978
2/3 45 9.689 0.192 0.052 0.981
2/3 50 9.737 0.318 1.108 0.846
3/3 5 10.054 0.538 0.372 0.766
3/3 10 9.860 0.504 0.406 0.761
3/3 15 9.880 0.439 0.141 0.951
3/3 20 9.843 0.258 0.158 0.970
3/3 25 9.866 0.178 0.058 0.977
3/3 30 9.853 0.165 0.092 0.955
3/3 35 9.884 0.189 0.086 0.982
3/3 40 9.892 0.221 0.057 0.983
3/3 45 9.824 0.144 0.065 0.978
3/3 50 9.878 0.194 0.050 0.983

The results show me that additional hidden layer will slow down the accuracy. With only one hidden layer we will reach higher than 90% quicker compare with others while ten hidden layers will be the slower to reach more than 90% accuracy.

I am thinking is that because dog and cat classification is only two outputs? Ia that why additional hidden layer will not be beneficial to out training? What do you think?