Max pooling is a downsampling operation used in convolutional neural networks (CNNs) that partitions feature maps into non-overlapping regions and retains only the maximum activation value from each region, effectively reducing spatial dimensionality while preserving the most salient features. In WiFi CSI sensing tasks such as people counting and queue detection, max pooling plays a critical role by suppressing noise, reducing computational overhead, and improving the model's robustness to minor spatial variations in CSI-derived representations such as recurrence plot images or amplitude-phase features. Common variants include global max pooling, which collapses an entire feature map to a single value and is often used before fully connected layers, and temporal max pooling, which is applied along the time dimension in sequential models to capture the most prominent signal patterns across time steps.
Dictionary term