Learning objectives
- • Reason about kernel size, stride, padding, and pooling
- • Explain why residual connections improve trainability
- • Connect architecture choices to latency and memory constraints
Deep Learning
CNN Design Patterns
Cover convolutions, receptive fields, and architecture choices that still surface in vision-heavy roles.
Recommended on day 2880 minutesIntermediate
Interview prompts
- • How does stride change the receptive field and compute cost?
- • Why was ResNet a turning point for deep vision models?
Prerequisites