LEARNING APPARATUS AND METHOD, PREDICTION APPARATUS AND METHOD, AND PROGRAM

• US 20110137834A1
• Filed: 11/24/2010
• Published: 06/09/2011
• Est. Priority Date: 12/04/2009
• Status: Active Grant

• Current Assignee: Sony Corporation (Sony Group Corp.)
• Original Assignee: Sony Corporation (Sony Group Corp.)
• Inventors: IDE, Naoki, Sabe, Kohtaro, Ito, Masato
• Granted Patent: US 8,572,008 B2
• Time in Patent Office: Days
• US Class Current: 706/12
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

A learning apparatus includes: a location acquiring section for acquiring time series data on locations of a user; a time acquiring section for acquiring time series data on times; and learning section for learning an activity model indicating an activity state of the user as a probabilistic state transition model, using the respective acquired time series data on the locations and the times as an input.

Claims

1. A learning apparatus comprising:

location acquiring means for acquiring time series data on locations of a user;

time acquiring means for acquiring time series data on times; and

learning means for learning an activity model indicating an activity state of the user as a probabilistic state transition model, using the respective acquired time series data on the locations and the times as an input.

2. The learning apparatus according to claim 1,

further comprising day acquiring means for acquiring time series data on days,

wherein the learning means learns the activity model indicating the activity state of the user, additionally using the time series data on the days as the input.

3. The learning apparatus according to claim 2,

further comprising at least one of traffic condition acquiring means for acquiring time series data on traffic conditions and weather condition acquiring means for acquiring time series data on weather conditions,

wherein the learning means learns the activity model indicating the activity state of the user, additionally using the time series data on at least one of the traffic conditions and the weather conditions as the input.

4. The learning apparatus according to claim 1, further comprising:

additional information acquiring means for acquiring time series data on additional information of one or more types other than the locations of the user and the times;

state series generating means for generating time series data on state nodes from the time series data on the locations and the times by inputting the acquired time series data on the locations and the times in the activity model of the user learned by the learning means and outputting the time series data on the state nodes of the activity model of the user;

selecting means for correspondingly storing the time series data on the additional information acquired by the additional information acquiring means and the time series data on the state nodes generated by the state series generating means, and for selecting only the time series data on the state nodes corresponding to the additional information at a current time point acquired by the additional information acquiring means from among the stored time series data in a case where the activity model of the user at the current time point is predicted using the activity model of the user learned by the learning means; and

re-calculating means for re-calculating a parameter for the activity model of the user by a statistical process, using the time series data on the state nodes selected by the selecting means.

5. A learning method performed by a learning apparatus which recognizes an activity model indicating an activity state of a user and learns the activity model of the user for use in a prediction apparatus which predicts a behavior of the user as a probabilistic state transition model, the method comprising the steps of:

acquiring time series data on locations of the user;

acquiring time series data on times; and

learning the activity model of the user as the probabilistic state transition model, using the respective acquired time series data on the locations and the times as an input.

6. A program which functions as the following means in a computer, the means comprising:

location acquiring means for acquiring time series data on locations of a user;

time acquiring means for acquiring time series data on times; and

learning means for learning an activity model indicating an activity state of the user as a probabilistic state transition model, using the respective acquired time series data on the locations and the times as an input.

7. A prediction apparatus comprising:

location acquiring means for acquiring time series data on locations of a user;

time acquiring means for acquiring time series data on times;

behavior recognizing means for recognizing a current location of the user using an activity model of the user, wherein the activity model indicates an activity state of the user and is obtained by being learned as a probabilistic state transition model, using the respective acquired time series data on the locations and the times as an input, by a learning apparatus;

behavior predicting means for predicting a possible path and a selection probability of the path from the current location of the user recognized by the behavior recognizing means; and

arrival time predicting means for predicting an arrival time and an arrival probability for arrival at a destination from the predicted path and selection probability.

8. The prediction apparatus according to claim 7,

further comprising day acquiring means for acquiring time series data on days,

wherein the learning apparatus learns the activity model of the user, additionally using the time series data on the days as the input, and

wherein the behavior recognizing means and the behavior predicting means use the activity model of the user obtained by performing the learning additionally using the time series data on the days as the input.

9. The prediction apparatus according to claim 7, further comprising:

additional information acquiring means for acquiring time series data on additional information of one or more types other than the locations of the user and the times;

selecting means for correspondingly storing time series data on state nodes generated by state series generating means of the learning apparatus from the time series data on the locations and the times by inputting the time series data on the locations and the times which is learning data in the activity model of the user which is learned and by outputting the time series data on the state nodes of the activity model of the user, and the time series data on the additional information at the time when the time series data on the locations and the times which is the learning data is acquired, and for selecting only the time series data on the state nodes corresponding to the additional information at a current time point acquired by the additional information acquiring means from among the stored time series data; and

re-calculating means for re-calculating a parameter for the activity model of the user by a statistical process, using the time series data on the state nodes selected by the selecting means.

10. The prediction apparatus according to claim 9,

wherein the learning apparatus learns the activity model of the user using time series data on movement speeds in addition to the locations of the user,

wherein the behavior recognizing means further recognizes a behavior state of the user at least including a movement state and a stationary state, using the activity model of the user obtained through the learning which additionally uses the time series data on the movement speeds of the user as the input, and

wherein the arrival time prediction means further predicts state nodes in which the movement state of the user becomes the stationary state as the destination.

11. A prediction method performed by a prediction apparatus which performs prediction using an activity model of a user in which a learning apparatus learns the activity model indicating an activity state of the user as a probabilistic state transition model, the method comprising the steps of:

acquiring time series data on locations of the user;

acquiring time series data on times; and

recognizing a current location of the user using the activity model of the user learned by the learning apparatus, using the respective acquired time series data on the locations and the times as an input;

predicting a possible path and a selection probability of the path from the recognized current location of the user; and

predicting an arrival time and an arrival probability for arrival at a destination from the predicted path and selection probability.

12. A program which functions as the following means in a computer, the means comprising:

location acquiring means for acquiring time series data on locations of a user;

time acquiring means for acquiring time series data on times;

behavior recognizing means for recognizing a current location of the user using an activity model of the user, wherein the activity model indicates an activity state of the user and is obtained by being learned as a probabilistic state transition model, using the respective acquired time series data on the locations and the times as an input, by a learning apparatus;

behavior predicting means for predicting a possible path and a selection probability of the path from the current location of the user recognized by the behavior recognizing means; and

arrival time predicting means for predicting an arrival time and an arrival probability for arrival at a destination from the predicted path and selection probability.

13. A learning apparatus comprising:

a location acquiring section which acquires time series data on locations of a user;

a time acquiring section which acquires time series data on times; and

a learning section which learns an activity model indicating an activity state of the user as a probabilistic state transition model, using the respective acquired time series data on the locations and the times as an input.

14. A prediction apparatus comprising:

a location acquiring section which acquires time series data on locations of a user;

a time acquiring section which acquires time series data on times;

a behavior recognizing section which recognizes a current location of the user using an activity model of the user, wherein the activity model indicates an activity state of the user and is obtained by being learned as a probabilistic state transition model, using the respective acquired time series data on the locations and the times as an input, by a learning apparatus;

a behavior predicting section which predicts a possible path and a selection probability of the path from the current location of the user recognized by the behavior recognizing section; and

an arrival time predicting section which predicts an arrival time for arrival at a destination and an arrival probability from the predicted path and selection probability.