Fully Connected Networks vs CNNs

Reported in interviews at Attentiv Groww Meesho Swiggy

Problem statement

How would you compare fully connected networks and CNNs for image classification? The team is considering dense and convolutional layers for color images. You need to understand their parameter counts and how they use spatial structure. Compare parameter sharing, spatial assumptions and expressive power.

Show hint

Compare how many weights each architecture needs for the same image.

Relevant roles: Deep Learning Engineer and ML Engineer.

Related free problems