Generalized method of moments — diverges in target from → Unsupervised risk estimation

Both are cited as partially-specified-model techniques in the same passage, but they partially specify toward different ends, and the pairing shows how flexible the umbrella idea is. The generalized method of moments partially specifies a distribution in order to identify a model's parameters — it answers "what is the true model." Unsupervised risk estimation partially specifies the distribution of a model's errors in order to estimate how badly a fixed, already-trained model is doing — it answers "how wrong is this model," without trying to say what the right model would be (concrete-problems, §"Partially specified models: method of moments, unsupervised risk estimation, causal identification, and limited-information maximum likelihood.", p. 18). The paper treats parameter identification and risk estimation as separate problems that happen to admit the same partial-specification strategy, rather than one being a special case of the other.