An overview of DiffEM, a method that uses diffusion models and expectation maximization to learn from corrupted data.