Inception — Understanding Multiple Parallel Convolutional Layers
Allowing the model to capture features at multiple scales simultaneously
In a previous post, we understood the architecture and built a sequential model using AlexNet as a reference. In this one, we will advance in complexity level and understand the concept of multiple parallel convolutional layers.
Why do we need multiple parallel convolutional layers?
Due to significant variability in the location and size of information within an image, multiple kernel sizes for the convolutional layers may be required. Larger kernels are used to capture globally distributed information, while smaller kernels are used to capture locally distributed information. Building a sequential model with sufficiently different kernel sizes results in a very deep model that is computationally expensive and susceptible to overfitting.
Multiple parallel convolutional layers are a design strategy where several convolutional layers operate in…
