Convolution, pooling and receptive field — Deep Learning & Neural Networks

Dense neural networks treat every input pixel as an independent feature — a 224×224 RGB image has 150,528 inputs, and a single Dense layer with 1,000 neurons would require 150 million parameters just for the first layer. This is computationally infeasible and statistically terrible: a dog's ear at position (10,20) and at position (80,120) would require completely separate weights to recognise as the same feature. Convolutional Neural Networks (CNNs) solve both problems with a single insight: local features repeat across space and should be detected with shared weights. A convolution filter is a small learnable matrix (3×3 or 5×5) that slides across the input, computing a dot product at each position — detecting the same edge, texture, or curve everywhere it appears. Pooling reduces spatial dimensions, building translation invariance and reducing computation. The receptive field quantifies how large a region of the input influences each output neuron — deeper networks have larger receptive fields and detect larger-scale features. Understanding convolution, pooling, and receptive fields is the prerequisite for understanding every image processing architecture from AlexNet to Vision Transformers.

Analogy🏏Cricket

🏏 Think of it like cricket: The Duckworth-Lewis-Stern (DLS) method, the ICC bowling economy metric, and the net run rate each measure team performance but optimise for completely different things — DLS cares about wickets-in-hand, economy cares about runs-per-over, NRR cares about tournament-wide margins. Using the wrong metric to evaluate a bowler is like using MSE for classification: technically computable but optimising the wrong thing entirely. Just as you would never rank a spinner by batting average, you should never use MSE when your model predicts probabilities. Just as focal DLS adjustments down-weight easy chases and amplify close finishes, focal loss down-weights easy examples and amplifies hard minority-class examples. The choice of scoring metric defines what excellence means — and so does the choice of loss function.