Dropout is a regularization technique used during the training of deep neural networks in which a random subset of neurons is temporarily "dropped out" (set to zero) at each training step, preventing the model from becoming overly reliant on any single neuron or feature. In WiFi CSI-based sensing tasks such as people counting and human activity recognition, dropout is critical for combating overfitting, particularly given the high dimensionality of CSI data and the relatively limited labeled datasets available in indoor environments. Common variants include standard dropout applied to fully connected layers, spatial dropout used within convolutional layers, and recurrent dropout applied within LSTM units to handle the temporal dependencies inherent in sequential CSI signal data.
Source Papers
- CSI-Based People Counting in WiFi Networks: Leveraging Occupancy Detection ↗ — CSI-Based People Counting in WiFi Networks: Leveraging Occup
- Efficient machine learning for Wi-Fi CSI-based human activity recognition using fast Monte Carlo based feature extraction ↗ — Efficient machine learning for Wi-Fi CSI-based human activit