Information processing apparatus, information processing method, and computer readable recording medium
The information processing apparatus 1 includes the input unit 101 that receives observation represented by a conjunction of atomic formulas, background knowledge represented by a set of logic formulas, and condition to be satisfied by a new rule represented by a logic formula; the rule candidate generation unit 102 that generates the new rule represented by a logic formula having a predicate, which is contained in the observation or the background knowledge, as an element; the abduction system unit 103 that executes abduction to derive best hypothesis by using the background knowledge, to which the new rule generated is added, and the observation as inputs; and the rule evaluation unit 104 that evaluates whether the new rule satisfies the condition to be satisfied by using the best hypothesis.
1 . An information processing apparatus comprising:
a memory having instructions stored therein;
at least one processor configured to execute the instructions, that when executed implement:
an input unit that receives observation represented by a conjunction of atomic formulas, background knowledge represented by a set of logic formulas, and condition to be satisfied by a new rule represented by a logic formula,
a rule candidate generation unit that generates the new rule represented by a logic formula having a predicate, which is contained in the observation or the background knowledge, as an element,
an abduction system unit that executes abduction to derive best hypothesis by using the background knowledge, to which the new rule generated is added, and the observation as inputs,
a rule evaluation unit that evaluates whether the new rule satisfies the condition to be satisfied by using the best hypothesis,
a hypothesis graph holding unit that holds a hypothesis graph generated prior to execution of abduction by the abduction system unit, the hypothesis graph having a graph which represents a set of possible hypotheses and has a node corresponding to a predicate, and
a rule pruning unit that determines to skip abduction to be executed with the background knowledge, to which the new rule is added, and the observation as inputs, by using a prior hypothesis graph generated when abduction is executed with the background knowledge, to which a rule having fewer logic formulas of conjunction than the new rule is added, and the observation as inputs.
2 . The information processing apparatus according to claim 1 , wherein
the rule pruning unit determines to skip abduction when a set of predicates, which are nodes contained in the prior hypothesis graph, does not contain a predicate that appears in a consequent of the new rule.
3 . The information processing apparatus according to claim 1 , wherein
the rule pruning unit determines to skip abduction when an evaluation result, evaluated by the rule evaluation unit using a best result derived based on the prior hypothesis graph, does not satisfy the condition and the logic formula of conjunction does not have a potential to be unified with any of the nodes of the prior hypothesis graph.
4 . An information processing method comprising:
receiving observation represented by a conjunction of atomic formulas, background knowledge represented by a set of logic formulas, and condition to be satisfied by a new rule represented by a logic formula,
generating the new rule represented by a logic formula having a predicate, which is contained in the observation or the background knowledge, as an element,
executing abduction to derive best hypothesis by using the background knowledge, to which the new rule generated is added, and the observation as inputs,
evaluating whether the new rule satisfies the condition to be satisfied by using the best hypothesis,
holding a hypothesis graph generated prior to execution of abduction, the hypothesis graph having a graph which represents a set of possible hypotheses and has a node corresponding to a predicate, and
determining to skip abduction to be executed with the background knowledge, to which the new rule is added, and the observation as inputs, by using a prior hypothesis graph generated when abduction is executed with the background knowledge, to which a rule having fewer logic formulas of conjunction than the new rule is added, and the observation as inputs.
5 . The information processing method according to claim 4 , wherein
when determining to skip abduction:
determining to skip abduction when a set of predicates, which are nodes contained in the prior hypothesis graph, does not contain a predicate that appears in a consequent of the new rule.
6 . The information processing method according to claim 4 , wherein
when determining to skip abduction:
determining to skip abduction when an evaluation result, evaluated using a best result derived based on the prior hypothesis graph, does not satisfy the condition and the logic formula of conjunction does not have a potential to be unified with any of the nodes of the prior hypothesis graph.
7 . A non-transitory computer-readable recording medium that includes a program recorded thereon, the program including instructions that cause a computer to execute:
receiving observation represented by a conjunction of atomic formulas, background knowledge represented by a set of logic formulas, and condition to be satisfied by a new rule represented by a logic formula,
generating the new rule represented by a logic formula having a predicate, which is contained in the observation or the background knowledge, as an element,
executing abduction to derive best hypothesis by using the background knowledge, to which the new rule generated is added, and the observation as inputs,
evaluating whether the new rule satisfies the condition to be satisfied by using the best hypothesis,
holding a hypothesis graph generated prior to execution of abduction, the hypothesis graph having a graph which represents a set of possible hypotheses and has a node corresponding to a predicate, and
determining to skip abduction to be executed with the background knowledge, to which the new rule is added, and the observation as inputs, by using a prior hypothesis graph generated when abduction is executed with the background knowledge, to which a rule having fewer logic formulas of conjunction than the new rule is added, and the observation as inputs.
8 . The non-transitory computer-readable recording medium according to claim 7 , wherein
when determining to skip abduction:
determining to skip abduction when a set of predicates, which are nodes contained in the prior hypothesis graph, does not contain a predicate that appears in a consequent of the new rule.
9 . The non-transitory computer-readable recording medium according to claim 7 , wherein
when determining to skip abduction:
determining to skip abduction when an evaluation result, evaluated using a best result derived based on the prior hypothesis graph, does not satisfy the condition and the logic formula of conjunction does not have a potential to be unified with any of the nodes of the prior hypothesis graph.