5.4. Which of the following is/are true about the Inception Architecture? 1 Mark

Using 1*1 convolution layers in Inception modules helped in dimension reduction and

reduction of the computational cost and training time.

Usage of 1*1 convolution layers, led to fewer feature maps but captured complex

features more optimally.

By increasing the number of units at each level's Inception modules (with dimension

reduction) and by stacking Inception layers we can make the model deeper, without

increasing the computational complexity.