Identifying Learning Rules From Neural Network Observables