Systems and methods for event prediction using schema networks
A system for event prediction using schema networks includes a first antecedent entity state that represents a first entity at a first time; a first consequent entity state that represents the first entity at a second time; a second antecedent entity state that represents a second entity at the first time; and a first schema factor that couples the first and second antecedent entity states to the first consequent entity state; wherein the first schema factor is configured to predict the first consequent entity state from the first and second antecedent entity states.
1. A system comprising one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:
storing a first schema factor network that is configured to generate a prediction of a goal entity state representing an entity at a later time from an antecedent entity state representing an entity at an earlier time;
performing a backward message passing procedure from the goal entity state to the antecedent entity state through the first schema factor network to determine a cause of a transition from the antecedent entity state to the goal entity state; and
using the cause of the transition to generate a plan comprising a sequence of one or more actions that achieve the goal entity state from the antecedent entity state.
2. The system of claim 1 , wherein backward message passing enables regression planning to determine the first value based on the second value.
3. The system of claim 1 , wherein a first schema factor of the first schema factor network comprises an OR factor, wherein the OR factor is connected to a first attribute and a second attribute of a second antecedent entity state, wherein the second antecedent entity state is associated with a second entity at the first time.
4. The system of claim 3 , wherein the first schema factor comprises a self-transition factor, wherein the self-transition factor is a variable that is associated with a probability that the first value will be different from the second value.
5. The system of claim 1 , wherein the a first schema factor of the first schema factor network comprises an AND factor, wherein the AND factor is connected to a first attribute and an action associated with the antecedent entity state.
6. The system of claim 5 , wherein the action is separate from the antecedent entity state.
7. The system of claim 1 , wherein the goal entity state and a second goal entity state are connected to a first schema factor of the schema factor network, wherein the first schema factor is connected to a third goal entity state, wherein the second goal entity state represents a second entity at the second time, and wherein the third goal entity state represents the first entity at a third time.
8. The system of claim 1 , wherein the antecedent entity state comprises a first attribute having a first value, and wherein the consequent entity state comprises the first attribute having a second value, and wherein the first attribute is a binary variable.
9. The system of claim 1 , wherein the antecedent entity state comprises a first attribute having a first value, and wherein the consequent entity state comprises the first attribute having a second value, and wherein the first attribute is a continuous variable discretized into multiple binary variables.
10. The system of claim 1 , wherein the antecedent entity state is determined by a visual network.
11. A method comprising:
storing a first schema factor network that is configured to generate a prediction of a goal entity state representing an entity at a later time from an antecedent entity state representing an entity at an earlier time;
performing a backward message passing procedure from the goal entity state to the antecedent entity state through the first schema factor network to determine a cause of a transition from the antecedent entity state to the goal entity state; and
using the cause of the transition to generate a plan comprising a sequence of one or more actions that achieve the goal entity state from the antecedent entity state.
12. The method of claim 11 , wherein backward message passing enables regression planning to determine the first value based on the second value.
13. The method of claim 11 , wherein a first schema factor of the first schema factor network comprises an OR factor, wherein the OR factor is connected to a first attribute and a second attribute of a second antecedent entity state, wherein the second antecedent entity state is associated with a second entity at the first time.
14. The method of claim 13 , wherein a first schema factor of the first schema factor network comprises an AND factor, wherein the AND factor is connected to a first attribute and an action associated with the antecedent entity state.
15. The method of claim 14 , wherein the action is separate from the antecedent entity state.
16. The method of claim 13 , wherein the first schema factor comprises a self-transition factor, wherein the self-transition factor is a variable that is associated with a probability that the first value will be different from the second value.
17. The method of claim 11 , wherein the goal entity state and a second goal entity state are connected to a first schema factor of the first schema factor network, wherein the first schema factor is connected to a third goal entity state, wherein the second goal entity state represents a second entity at the second time, and wherein the third goal entity state represents the first entity at a third time.
18. The method of claim 11 , wherein the antecedent entity state comprises a first attribute having a first value, and wherein the consequent entity state comprises the first attribute having a second value, and wherein the first attribute is a binary variable.
19. The method of claim 11 , wherein the antecedent entity state comprises a first attribute having a first value, and wherein the consequent entity state comprises the first attribute having a second value, and wherein the first attribute is a continuous variable discretized into multiple binary variables.
20. The method of claim 11 , wherein the antecedent entity state is determined by a visual network.
21. One or more non-transitory computer storage media encoded with computer program instructions that when executed by one or more computers cause the one or more computers to perform operations comprising:
storing a first schema network that is configured to generate a prediction of a goal entity state representing an entity at a later time from an antecedent entity state representing an entity at an earlier time;
performing a backward message passing procedure from the goal entity state to the antecedent entity state through the first schema factor network to determine a cause of a transition from the antecedent entity state to the goal entity state; and
using the cause of the transition to generate a plan comprising a sequence of one or more actions that achieve the goal entity state from the antecedent entity state.