STATE ESTIMATION APPARATUS, STATE ESTIMATION METHOD, AND INTEGRATED CIRCUIT

• US 20170278006A1
• Filed: 03/24/2017
• Published: 09/28/2017
• Est. Priority Date: 03/25/2016
• Status: Active Grant

• Current Assignee: Megachips Corporation
• Original Assignee: Megachips Corporation
• Inventors: HASEGAWA, Hiromu
• Granted Patent: US 10,586,154 B2
• Time in Patent Office: Days
• CPC Class Codes: G06N 3/088 Non-supervised learning, e.g. competitive learning

Abstract

Provided is a state determination apparatus that appropriately performs pattern classification processing and/or pattern determination processing even when a map generated by the SOM technique includes discontinuous image regions (e.g., split image regions). In the state determination apparatus, the matching processing unit obtains adaptability data indicating a correlation degree between template data indicating a predetermined state and the SOM output data. The state determination unit determines a state of an input data. This allows for appropriately performing pattern classification processing and/or pattern determination processing even when a map generated by the SOM technique includes discontinuous image regions (e.g., split image regions).

Claims

1. A state determination apparatus comprising:

feature vector obtaining circuitry configured to obtain feature vector data from measured data obtained by measuring an event with an unknown state;

mapping conversion circuitry configured to obtain SOM output data by mapping the feature vector data into a space whose dimension differs from a dimension of the feature vector data;

matching processing circuitry configured to obtain adaptability data indicating a correlation degree between template data indicating a predetermined state and the SOM output data; and

state determination circuitry configured to determine a state indicated by the measured data based on the adaptability data.

2. The state determination apparatus according to claim 1, wherein

the matching processing circuitry obtains M (M is a natural number) pieces of adaptability data D_f(1) to D_f(M), which respectively indicate the correlation degree with respect to M pieces of template data Tmpl(1) to Tmpl(M), and

the state determination circuitry specifies the template data corresponding to the adaptability data whose correlation degree is the highest of the adaptability data D_f(1) to D_f(M), and determines that a state corresponding to the specified template data is the state indicated by the measured data.

3. The state determination apparatus according to claim 2, wherein

the matching processing circuitry calculates inner products using M (M is a natural number) pieces of the template data and the SOM output data and obtains the calculated inner products as the adaptability data D_f(1) to D_f(M), and

the state determination circuitry specifies the template data corresponding to the adaptability data whose correlation degree is the highest of the adaptability data D_f(1) to D_f(M) and determines that a state corresponding to the specified template data is the state indicated by the measured data.

4. The state determination apparatus according to claim 2, wherein

the matching processing circuitry calculates Euclidean distances using M (M is a natural number) pieces of the template data and the SOM output data and obtains the calculated Euclidean distances as the adaptability data D_f(1) to D_f(M), and

the state determination circuitry specifies the template data corresponding to the adaptability data whose correlation degree is the highest of the adaptability data D_f(1) to D_f(M) and determines that a state corresponding to the specified template data is the state indicated by the measured data.

10. The state determination apparatus according to claim 1, wherein

the template data is composed of two-dimensional elements, and

a sum of all values of the two-dimensional elements is “

1”

.

11. The state determination apparatus according to claim 1, wherein

the template data is composed of two-dimensional elements,

the template data is created based on N pieces of SOM output data obtained during a period from timing t=1 to timing t=N (N is a natural number) in which an identical state continues, and

assuming that the template data is composed of m×

n (m and n are natural numbers) pieces of data, the template data is created by setting a value g_ij to element data at a second-dimensional coordinate (i, j) (i and j are natural numbers satisfying 1≦

i≦

m and 1≦

j≦

n) of the template, the value g_ij being obtained by

12. The state determination apparatus according to claim 1, wherein

the template data is composed of two-dimensional elements,

the template data is created based on N pieces of SOM output data obtained during a period from timing t=1 to timing t=N (N is a natural number) in which an identical state continues, and

assuming that (1) the template data is composed of m×

n (m and n are natural numbers) pieces of data, (2) g_ij is element data at a second-dimensional coordinate (i, j) (i and j are natural numbers satisfying 1≦

i≦

m and 1≦

j≦

n) of the template data, and (3) D_ij(t) is element data at a second-dimensional coordinate (i, j) of the SOM output data at timing t, and the template data is created by setting an average value obtained by calculating an average value of the element data D_ij(t) satisfying D_ij(t)≧

Th1 to a value g_ij at a second-dimensional coordinate (i, j) of the template data.

13. The state determination apparatus according to claim 1, wherein

the template data is composed of two-dimensional elements,

the template data is created based on N pieces of SOM output data obtained during a period from timing t=1 to timing t=N (N is a natural number) in which an identical state continues, and

assuming that (1) the template data is composed of m×

n (m and n are natural numbers) pieces of data, (2) g_ij is element data at a second-dimensional coordinate (i, j) (i and j are natural numbers satisfying 1≦

i≦

m and 1≦

j≦

n) of the template data, and (3) D_ij(t) is element data at a second-dimensional coordinate (i, j) of the SOM output data at timing t, and the template data is created by setting a value g_ij to element data at a second-dimensional coordinate (i, j) (i and j are natural numbers satisfying 1≦

i≦

m and 1≦

j≦

n) of the template, the value g_ij being obtained by calculating an average value of values D_ij(t) of the SOM output data at the coordinate (i, j) whose corresponding variance value v_ij is obtained by

14. The state determination apparatus according to claim 1, wherein

the template data is composed of two-dimensional elements,

the template data is created based on N pieces of SOM output data obtained during a period from timing t=1 to timing t=N (N is a natural number) in which an identical state continues, and

assuming that (1) the template data is composed of m×

n (m and n are natural numbers) pieces of data, (2) g_ij is element data at a second-dimensional coordinate (i, j) (i and j are natural numbers satisfying 1≦

i≦

m and 1≦

j≦

n) of the template data, and (3) D_ij(t) is element data at a second-dimensional coordinate (i, j) of the SOM output data at timing t, and the template data is created by setting a value g_ij to element data at a second-dimensional coordinate (i, j) (i and j are natural numbers satisfying 1≦

i≦

m and 1≦

j≦

n) of the template, the value g_ij being obtained by calculating an average value of values D_ij(t) of the SOM output data at the coordinate (i, j) whose corresponding variance value v_ij is obtained by

15. The state determination apparatus according to claim 1, wherein

the template data is composed of two-dimensional elements,

the template data is created based on N pieces of SOM output data obtained during a period from timing t=1 to timing t=N (N is a natural number) in which an identical state continues, and

assuming that (1) the template data is composed of m×

n (m and n are natural numbers) pieces of data, (2) g_ij is element data at a second-dimensional coordinate (i, j) (i and j are natural numbers satisfying 1≦

i≦

m and 1≦

j≦

n) of the template data, (3) D_ij(t) is element data at a second-dimensional coordinate (i, j) of the SOM output data at timing t, and (4) σ

_ij is a standard deviation of element data at the coordinate (i, j) obtained by

5. A state determination apparatus comprising:

feature vector obtaining circuitry configured to obtain feature vector data from measured data obtained by measuring an event with an unknown state;

normalization circuitry configured to obtain a norm of the feature vector data and obtain normalized feature vector data by normalizing the feature vector data;

mapping conversion circuitry configured to obtain SOM output data by mapping the normalized feature vector data into a space whose dimension differs from a dimension of the normalized feature vector data;

matching processing circuitry configured to obtain adaptability data indicating a correlation degree between template data indicating a predetermined state and the SOM output data;

activity value obtaining circuitry configured to obtain an activity value indicating a probability that the norm obtained by the normalization circuitry corresponds to a norm of the feature vector data obtained in the predetermined state; and

state determination circuitry configured to determine a state indicated by the measured data based on the adaptability data and the activity value.

6. The state determination apparatus according to claim 5, wherein

the activity value obtaining circuitry obtains the activity value for a state k (k is a natural number satisfying 1≦

k≦

M, and M is a natural number) through processing corresponding to

7. The state determination apparatus according to claim 5, wherein

the matching processing circuitry obtains M (M is a natural number) pieces of adaptability data D_f(1) to D_f(M), which respectively indicate the correlation degree with respect to M pieces of template data Tmpl(1) to Tmpl(M),

the activity value obtaining circuitry obtains activity values D_act(1) to D_act(M), which respectively corresponds to a first state to M-th state, and

the state determination circuitry obtains a determination evaluation value D_est(k) for k-th state (k is a natural number satisfying 1≦

k≦

M, and M is a natural number) based on the adaptability data D_f(k) for the k-th state and the activity value D_act(k) for the k-th state, and determines a state indicated by the measured data based on the obtained determination evaluation value D_est(k) for the k-th state.

8. The state determination apparatus according to claim 7, wherein

the matching processing circuitry calculates inner products using M (M is a natural number) pieces of the template data and the SOM output data and obtains the calculated inner products as the adaptability data D_f(1) to D_f(M), and

the state determination circuitry obtains the determination evaluation value D_est(k) for the k-th state through processing corresponding to

D_est(k)=h(D_f(k),D_act(k))

h(x,y):

a function of variables x and y, and

determines that a state corresponding to the largest determination evaluation value of the determination evaluation values D_est(1) to D_est(M) is the state indicated by the measured data.

9. The state determination apparatus according to claim 7, wherein

the matching processing circuitry calculates Euclidean distances using M (M is a natural number) pieces of the template data and the SOM output data and obtains the calculated Euclidean distances as the adaptability data D_f(1) to D_f(M), and

the state determination circuitry obtains the determination evaluation value D_est(k) for the k-th state through processing corresponding to

D_est(k)=h(F1(D_f(k)),D_act(k))

where h(x,y) is a function of variables x and y, and F1(x) is a monotonically decreasing function with respect to x and the value of the function F1(x) is a positive value or zero, and determines that a state corresponding to the largest determination evaluation value of the determination evaluation values D_est(1) to D_est(M) is the state indicated by the measured data.