Systems and methods to customize student instruction

• US 10,290,221 B2
• Filed: 04/29/2013
• Issued: 05/14/2019
• Est. Priority Date: 04/27/2012
• Status: Active Grant

• Current Assignee: Aptima Inc.
• Original Assignee: Aptima Inc.
• Inventors: Stacy, E. Webb, Dean, Courtney, Carlin, Alan, Dumond, Danielle
• Primary Examiner(s): Utama, Robert J
• Application Number:
  US13/872,519
• Publication Number:
  US 20130288222A1
• Time in Patent Office: 2,206 Days
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G09B 5/00 Electrically-operated educational appliances working with questions and answers G09B7/00; simulators G09B9/00; advertising or displaying in general G09F
• CPC Class Codes: G09B 7/00 Electrically-operated teaching apparatus or devices working with questions and answers mechanically operated G09B3/00; computing arrangements G06F

Abstract

A computer implemented systems and methods for determining an action for a user within a learning domain are disclosed, some embodiments of the methods comprise defining an initial learning model of a learning domain, determining an initial user state of the user, determining an initial user action from at least one learning domain action with the initial learning model, receiving a user observation of the user after the user executes the initial user action, determining an updated user state with the initial learning model given the updated user observation and determining a subsequent user action from the at least one learning domain action. Some embodiments utilize a Partially Observable Markov Model (POMDP) as the learning model.

Claims

1. A computer implemented method for determining an action for a user within a learning domain, the method comprising:

defining an initial learning model of a learning domain comprising;

a plurality of learning domain states,

at least one learning domain action,

at least one domain learning domain state transition,

and at least one learning domain observation;

determining an initial user state of the user;

the initial user state comprising one of the plurality of learning domain states;

determining an initial user action from the at least one learning domain action with the initial learning model given the initial user state as the one of the plurality of learning domain states;

receiving a user observation of the user after the user executes the initial user action;

the user observation comprising one of the at least one learning domain observations;

determining, with a processor, an updated user state from the plurality of learning domain states with the initial learning model given the user observation; and

determining a subsequent user action from the at least one learning domain action.

2. The computer implemented method of claim 1 wherein:

the initial learning model comprises a Partially Observable Markov Decision Process (POMDP); and

the at least one domain learning domain state transition comprises a learning domain state transition probability table of P (s′

|s, a) where s is the initial user state, s′

is the updated user state and a is the at least one learning domain action.

3. The computer implemented method of claim 2 further comprising automatically determining the learning domain state transition probability table utilizing a Hidden Markov Model (HMM) function where an input to the HMM function comprises a probability that the initial user state will change to the updated user state, a probability that the learning domain observation is obtained while a user state of the user is the updated user state and a probability of the initial user state.

4. The computer implemented method of claim 3 wherein the HMM function comprises a state transition probability distribution, a conditional observation probability distribution, and a probability of the initial user state whereby the state transition probability distribution defines the at least one domain learning domain state transition for the POMDP given the at least one learning domain observation defined by the conditional observation probability distribution for the initial user state defined by the probability of the initial user state.

5. The computer implemented method of claim 1 further comprising defining the plurality of learning domain states, the at least one learning domain action and the at least one learning domain observation from a learning domain ontology.

6. The computer implemented method of claim 5 wherein:

defining an initial learning model of a learning domain further comprises automatically determining the learning domain ontology from a learning domain data set comprising a plurality of learning domain problems and a plurality of learning domain observations; and

the plurality of learning domain problems comprise a plurality of experiential learning modules presented to the user by a computer based adaptive trainer.

7. The computer implemented method of claim 6 wherein:

automatically determining the learning domain ontology from a learning domain data set further comprises utilizing a Principal Component Analysis (PCA) algorithm whereby an input is the plurality of learning domain problems and the plurality of learning domain observations and an output is at least one of the plurality of learning domain states.

8. The computer implemented method of claim 7 wherein:

the initial learning model comprises a Partially Observable Markov Decision Process (POMDP); and

the at least one of the plurality of learning domain states defines the plurality of learning domain states for the POMDP.

9. The computer implemented method of claim 1 wherein:

the plurality of learning domain states comprises a competency measure; and

defining the initial learning model of a learning domain further comprises automatically determining the competency measure.

10. The computer implemented method of claim 9 wherein defining the initial learning model of a learning domain further comprises automatically determining the competency measure further comprises utilizing an Item Response Theory (IRT) algorithm comprising a conditional probability of getting at least one item correct given a latent proficiency score of the user and a difficulty level of the at least one item.

11. The computer implemented method of claim 10 wherein:

the initial learning model comprises a Partially Observable Markov Decision Process (POMDP); and

a conditional probability of getting the at least one item correct given a latent proficiency score of the user and a difficulty level of the at least one item defines the at least one learning domain observation for the POMDP.

12. The computer implemented method of claim 3 wherein:

automatically determining a learning domain ontology from a learning domain data set further comprises utilizing a Principal Component Analysis (PCA) algorithm whereby the input is a plurality of learning domain problems and a plurality of learning domain observations and an output is at least one of the plurality of learning domain states;

the plurality of learning domain states defines the plurality of learning domain states for the POMDP;

defining the initial learning model of a learning domain further comprises automatically determining a competency measure utilizing an Item Response Theory (IRT) algorithm comprising a conditional probability of getting at least one item correct given a latent proficiency score of the user and a difficulty level of the at least one item; and

the a conditional probability of getting the at least one item correct given a latent proficiency score of the user and a difficulty level of the at least one item defines the at least one learning domain observation for the POMDP.

19. The computer implemented method of claim 1 wherein:

the initial user action comprises at least one of a test, a quiz or a problem; and

the at least one learning domain observation comprises a user'"'"'s score on the test, the quiz or the problem.

20. The computer implemented method of claim 1 wherein:

the at least one learning domain action comprises at least one initial learning module;

the initial user action comprises presenting the initial learning module to the user; and

the at least one learning domain observation comprises a physiological measure of the user when the user executes the initial user action.

13. A learning model system for determining an action for a user within a learning domain, the learning model system comprising a computer system including one or multiple processors configured to perform the functions of:

defining an initial learning model of a learning domain comprising;

a plurality of learning domain states,

at least one learning domain action,

at least one domain learning domain state transition,

and at least one learning domain observation;

determining an initial user state of the user;

the initial user state comprising one of the plurality of learning domain states;

determining an initial user action from the at least one learning domain action with the initial learning model given the initial user state as the one of the plurality of learning domain states;

receiving a user observation of the user after the user executes the initial user action;

the user observation comprising one of the at least one learning domain observations;

determining, with a processor, an updated user state from one of the plurality of learning domain states with the initial learning model given the user observation; and

determining a subsequent user action from the at least one learning domain action.

14. The learning model system of claim 13 wherein:

the initial learning model comprises a Partially Observable Markov Decision Process (POMDP); and

the at least one domain learning domain state transition comprises a learning domain state transition probability table of P (s′

|s, a) where s is the initial user state, s′

is the updated user state and a is the at least one learning domain action.

15. The learning model system of claim 14 wherein the one or more processors are further configured to perform the functions of automatically determining the learning domain state transition probability table utilizing a Hidden Markov Model (HMM) function where an input to the HMM function comprises a probability that the initial user state will change to the updated user state, a probability that the learning domain observation is obtained while a user state of the user is the updated user state and a probability of the initial user state.

16. The learning model system of claim 13 wherein:

the initial learning model comprises a Partially Observable Markov Decision Process (POMDP);

the one or more processors are further configured to perform the functions of;

automatically determining a learning domain ontology from a learning domain data set utilizing a Principal Component Analysis (PCA) algorithm whereby an input is a plurality of learning domain problems and a plurality of learning domain observations and an output is at least one of the plurality of learning domain states; and

the plurality of learning domain states defines the plurality of learning domain states for the POMDP.

17. The learning model system of claim 13 wherein:

the plurality of learning domain states comprises a competency measure;

the one or more processors are further configured to perform the functions of;

defining the initial learning model of a learning domain further comprises automatically determining the competency measure utilizing an Item Response Theory (IRT) algorithm comprising a conditional probability of getting at least one item correct given a latent proficiency score of the user and a difficulty level of the at least one item;

the initial learning model comprises a Partially Observable Markov Decision Process (POMDP); and

the conditional probability of getting the at least one item correct given a latent proficiency score of the user and a difficulty level of the at least one item defines the at least one learning domain observation for the POMDP.

21. The learning model system of claim 13 wherein:

the learning model system is a computer based adaptive trainer;

the user is a student communicating with the computer based adaptive trainer;

the plurality of learning domain states comprise a plurality of competency measures to define at least the initial user state and the updated user state of the user;

the at least one learning domain action comprises at least one learning module presented to the user by the computer based adaptive trainer;

the at least one domain learning domain state transition comprises a learning domain state transition probability table of P (s′

|s, a) where s is the initial user state, s′

is the updated user state and a is the at least one learning domain action;

the at least one learning domain observation is an observation of a performance of the user on the learning module presented to the user; and

the subsequent user action from the at least one learning domain action comprises a learning domain action determined specifically for the user based on the updated user state.

18. A computer program product for a learning model system comprising a non-transitory computer readable storage medium having a computer readable program code embodied therein, said computer readable program code configured to be executed to implement a method determining an action for a user within a learning domain, comprising:

defining an initial learning model of a learning domain comprising;

a plurality of learning domain states,

at least one learning domain action,

at least one domain learning domain state transition,

and at least one learning domain observation;

determining an initial user state of the user;

the initial user state comprising one of the plurality of learning domain states;

the plurality of learning domain states comprise a competency measure;

determining an initial user action from the at least one learning domain action with the initial learning model given the initial user state as the one of the plurality of learning domain states;

receiving a user observation of the user after the user executes the initial user action;

the user observation comprising one of the at least one learning domain observations;

determining, with a processor, an updated user state from one of the plurality of learning domain states with the initial learning model given the user observation; and

determining a subsequent user action from the at least one learning domain action.