Expectation-Maximization (EM) is an iterative statistical optimization algorithm used to estimate unknown parameters in probabilistic models, particularly when data is incomplete, missing, or involves latent (hidden) variables; it alternates between an Expectation step, which computes the expected value of the log-likelihood given current parameter estimates, and a Maximization step, which updates parameters to maximize that expectation. In the context of crowd and pedestrian sensing research, EM is relevant for tasks such as fitting Gaussian mixture models to crowd density distributions, inferring hidden state variables in agent-based models, and performing data assimilation by reconciling noisy observations with model priors. Key variants include the hard-assignment K-means analog, variational EM for approximate inference in complex models, and online or incremental EM formulations suited to streaming or large-scale data scenarios encountered in real-time crowd monitoring.
Source Papers
- Data Assimilation for Agent-Based Models ↗ — Data Assimilation for Agent-Based Models
- Device-Free Passive Identity Identification via WiFi Signals ↗ — Device-Free Passive Identity Identification via WiFi Signals
- Enabling ISAC on Low-Cost Devices via Spatial-Channel Estimation With a Single-RF Chain ↗ — Enabling ISAC on Low-Cost Devices via Spatial-Channel Estima
- Recent trends in crowd analysis: A review ↗ — Recent trends in crowd analysis: A review