Learning method, learning apparatus and program
A learning method to be executed by a computer according to one embodiment includes of acquiring event history information representing a history of a predetermined event; and training, by using the acquired event history information, parameters of an intensity function in which a trigger function is set to be a function represented by a composite function of a first function and a predetermined second function; and a derivative of the first function, the first function being represented by a neural network that models a temporal change in influence of the event.
1 . A computer-implemented method for training a model, comprising:
acquiring event history information representing a history of a predetermined event, wherein the event history information comprises a set of timestamps indicating occurrences of a past event of a predetermined type over time, and the model comprises an intensity function; and
training, by using the acquired event history information, parameters of the intensity function of the model by setting a trigger function as a part of the intensity function,
wherein the trigger function comprises a composite function of a first function and a predetermined second function,
the trigger function is represented by a product of g(f(Δ i )) and the derivative of the first function f with respect to Δ i , where Δ i =t−t i denotes a difference between a time point t i of occurrence of the past event in the past and a current time point t, f denotes the first function, and g denotes the second function,
the first function comprises a neural network that describes a temporal change of influence of the occurrences of the past event of the predetermined type,
the trigger function further comprises a derivative of the first function, and
the trigger function represents a part of describing a likelihood of the occurrences of the past event over time according to the event history information.
2 . The computer-implemented method according to claim 1 , wherein the training further comprises minimizing a likelihood comprising the first function and the derivative of the first function.
3 . The computer-implemented method according to claim 1 , wherein the first function f represents a function that integrates a function a in an interval [t j , t], where the function a represents the temporal change of the influence of the occurrences of the past event of the predetermined type.
4 . The computer-implemented method according to claim 1 , wherein the second function represents an exponential decay function, a Weibull distribution, or a gamma distribution.
5 . A learning apparatus, comprising:
a processor; and
a memory storing program instructions that cause the processor to:
acquire event history information representing a history of a predetermined event, wherein the event history information comprises a set of timestamps indicating occurrences of a past event of a predetermined type over time, and the model comprises an intensity function; and
train, by using the acquired event history information, parameters of the intensity function of the model by setting a trigger function as a part of the intensity function,
wherein the trigger function comprises a composite function of a first function and a predetermined second function,
the trigger function is represented by a product of g(f(Δ i ) and the derivative of the first function f with respect to Δ i , where Δ i =t−t i denotes a difference between a time point t i of occurrence of the past event in the past and a current time point t, f denotes the first function, and g denotes the second function,
the first function comprises a neural network that describes a temporal change of influence of the occurrences of the past event of the predetermined type,
the trigger function further comprises a derivative of the first function, and
the trigger function represents a part of describing a likelihood of the occurrences of the past event over time according to the event history information.
6 . A non-transitory computer-readable storage medium that stores therein a program for causing a computer to execute the computer-implemented method according to claim 1 .
7 . The computer-implemented method according to claim 1 , wherein the model after trained predicts, based on input time, an occurrence of an event of the predetermined type.
8 . The computer-implemented method according to claim 1 , further comprising:
predicting, based on input time, by the model after trained, an occurrence of an event of the predetermined type.
9 . The computer-implemented method according to claim 1 , wherein the event history information further comprises user information, and
the model after trained predicts, based on input time and input user information, an occurrence of an event of the predetermined type by a user as described by the user information.
10 . The computer-implemented method according to claim 1 , wherein the predetermined type represents inflection of an infectious disease.
11 . The computer-implemented method according to claim 1 , wherein the predetermined type represents an act of spreading content on a social network service.
12 . The learning apparatus according to claim 5 , wherein the model after trained predicts, based on input time, an occurrence of an event of the predetermined type.
13 . The learning apparatus according to claim 5 , the program instructions further cause the processor to:
predict, based on input time, by the model after trained, an occurrence of an event of the predetermined type.
14 . The learning apparatus according to claim 5 , wherein the event history information further comprises user information, and
the model after trained predicts, based on input time and input user information, an occurrence of an event of the predetermined type by a user as described by the user information.
15 . The learning apparatus according to claim 5 , wherein the predetermined type represents inflection of an infectious disease.
16 . The learning apparatus according to claim 5 , wherein the predetermined type represents an act of spreading content on a social network service.