References

The complexity for abducing candidates could be exponential in the worst case. A.4.2 NeuralNetwork&Hyperparameters All compared methods share the same neural network structure for the same dataset. The default hyperparameters have low performance in the experiments. We fine-tune the hyperparameters and use a learningrateof10 3 andasearchstepof50. There are eight kinds ofsentencing element labels, includingrecidivism,confession,surrender, juvenile, forgiveness, no loss, pickpocket, burglary.