System and Method for Data Fusion with Adaptive Learning

• US 20120215735A1
• Filed: 02/18/2011
• Published: 08/23/2012
• Est. Priority Date: 02/18/2011
• Status: Active Grant

• Current Assignee: LARK Technologies, Inc. (Abbvie Incorporated)
• Original Assignee: LARK Technologies, Inc. (Abbvie Incorporated)
• Inventors: Abielmona, Rami, Di Nardo, George
• Granted Patent: US 8,478,711 B2
• Time in Patent Office: Days
• US Class Current: 706/52
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

There are disclosed techniques for performing data fusion. In one embodiment, the technique comprises: (a) receiving an input; (b) processing the input in accordance with parameters of a stochastic model to derive a state of a stochastic process of the stochastic model, the parameters of the stochastic model previously established during a training period; (c) obtaining a predicted next state of the stochastic process; (d) receiving a next input and processing the next input in accordance with the parameters of the stochastic model to derive a next state of the stochastic process; (e) comparing either (i) the next state and the predicted next state, or (ii) the next input and a predicted next input corresponding to the predicted next state; and (f) if there is a discrepancy between that compared in (e), then using the next input to modify the parameters of the stochastic model.

Claims

1. A method for performing the fusion of data in a data fusion engine, the method comprising:

(a) receiving an input at a data fuser in the data fusion engine;

(b) processing the input in the data fuser in accordance with parameters of a stochastic model to derive a state of a stochastic process of the stochastic model, the parameters of the stochastic model previously established during a training period;

(c) obtaining a predicted next state of the stochastic process;

(d) receiving a next input at the data fuser and processing the next input in the data fuser in accordance with the parameters of the stochastic model to derive a next state of the stochastic process;

(e) comparing either (i) the next state and the predicted next state, or (ii) the next input and a predicted next input corresponding to the predicted next state and derived from the predicted next state;

(f) if there is a discrepancy between that compared in (e), then using the next input to modify the parameters of the stochastic model.

2. The method of claim 1 wherein the stochastic model is a hidden markov model, wherein the stochastic process of the stochastic model is a hidden markov process of the hidden markov model, wherein said parameters of the stochastic model are parameters of the hidden markov model, and the parameters of the hidden markov model include at least one of:

hidden states of the hidden markov model, observations for hidden states of the hidden markov model, a state transition matrix, and emission probabilities for the observations of the hidden states.

3. The method of claim 2 wherein the data fusion engine comprises a plurality of cells arranged in a layered hierarchical structure having a bottom layer, a top layer, and m middle layers;

wherein an input of each cell in the top layer and the middle layers is obtained from an output of at least one cell in an adjacent lower layer; and

wherein operations (a) to (f) are performed in one or more cells of the plurality of cells.

4. The method of claim 3 wherein operations (a) to (f) are performed in each cell of the plurality of cells, and wherein the output of the at least one cell in each adjacent lower layer comprises an assumed state of a hidden markov process of the at least one cell.

5. The method of claim 4 further comprising:

each cell in the top layer and the middle layers deriving from its predicted next state;

a predicted next state of a corresponding cell in its adjacent lower layer; and

each cell in the top layer and the middle layers transmitting to the corresponding cell the predicted next state of the corresponding cell.

6. The method of claim 4 wherein said using the next input to modify the parameters of the hidden markov model comprises:

providing the next input as an input to a Baum-Welch algorithm to generate modified parameters of the hidden markov model.

7. The method of claim 3 wherein said obtaining the predicted next state comprises:

obtaining from the state transition matrix the predicted next state by selecting from the state transition matrix a most likely state at a next step of the hidden markov process given said state.

8. A data fusion engine for performing the fusion of data comprising:

a data fuser implementing a stochastic model;

a memory for storing parameters of the stochastic model;

a comparator; and

a parameter modifier for modifying parameters of the stochastic model;

the data fusion engine for;

(a) receiving an input at the data fuser;

(b) processing the input in the data fuser in accordance with parameters of the stochastic model to derive a state of the stochastic process, the parameters of the stochastic model previously established during a training period;

(c) obtaining a predicted next state of the stochastic process;

(d) receiving a next input at the data fuser and processing the next input in the data fuser in accordance with the parameters of the stochastic model to derive a next state of the stochastic process;

(e) comparing in the comparator either (i) the next state and the predicted next state, or (ii) the next input and a predicted next input corresponding to the predicted next state and derived from the predicted next state;

(f) if there is a discrepancy between that compared in (e), the parameter modifier using the next input to modify the parameters of the stochastic model.

9. The data fusion engine of claim 8 wherein the stochastic model is a hidden markov model, wherein the stochastic process of the stochastic model is a hidden markov process of the hidden markov model, wherein said parameters of the stochastic model are parameters of the hidden markov model, and the parameters of the hidden markov model include at least one of:

hidden states of the hidden markov model, observations for hidden states of the hidden markov model, a state transition matrix, and emission probabilities for the observations of the hidden states.

10. The data fusion engine of claim 9 wherein the data fusion engine comprises a plurality of cells arranged in a layered hierarchical structure having a bottom layer, a top layer, and m middle layers;

wherein an input of each cell in the top layer and the middle layers is obtained from an output of at least one cell in an adjacent lower layer; and

wherein operations (a) to (f) are performed in one or more cells of the plurality of cells.

11. The data fusion engine of claim 10 wherein the data fusion engine is configured to perform operations (a) to (f) in each cell of the plurality of cells, and wherein the output of the at least one cell in each adjacent lower layer comprises an assumed state of a hidden markov process of the at least one cell.

12. The data fusion engine of claim 11, wherein the data fusion engine is further configured for:

each cell in the top layer and the middle layers deriving from its predicted next state;

a predicted next state of a corresponding cell in its adjacent lower layer; and

each cell in the top layer and the middle layers transmitting to the corresponding cell the predicted next state of the corresponding cell.

13. The data fusion engine of claim 10 wherein said obtaining the predicted next state comprises:

obtaining from the state transition matrix the predicted next state by selecting from the state transition matrix a most likely state at a next step of the hidden markov process given said state.

20. A territorial security system comprising:

an input for receiving data from a plurality of data sources;

the data fusion engine of claim 8;

an architecture for processing fused data from the data fusion engine to obtain an intelligence product.

14. A computer readable medium having stored thereon computer readable instructions for performing the fusion of data in a data fusion engine, the computer readable instructions including instructions for performing operations comprising:

(a) receiving an input at a data fuser in the data fusion engine;

(b) processing the input in the data fuser in accordance with parameters of a stochastic model to derive a state of a stochastic process of the stochastic model, the parameters of the stochastic model previously established during a training period;

(c) obtaining a predicted next state of the stochastic process;

(d) receiving a next input at the data fuser and processing the next input in the data fuser in accordance with the parameters of the stochastic model to derive a next state of the stochastic process;

(e) comparing either (i) the next state and the predicted next state, or (ii) the next input and a predicted next input corresponding to the predicted next state and derived from the predicted next state;

(f) if there is a discrepancy between that compared in (e), then using the next input to modify the parameters of the stochastic model.

15. The computer readable medium of claim 14 wherein the stochastic model is a hidden markov model, wherein the stochastic process of the stochastic model is a hidden markov process of the hidden markov model, wherein said parameters of the stochastic model are parameters of the hidden markov model, and the parameters of the hidden markov model include at least one of:

hidden states of the hidden markov model, observations for hidden states of the hidden markov model, a state transition matrix, and emission probabilities for the observations of the hidden states.

16. The computer readable medium of claim 15 wherein the data fusion engine comprises a plurality of cells arranged in a layered hierarchical structure having a bottom layer, a top layer, and m middle layers;

wherein an input of each cell in the top layer and the middle layers is obtained from an output of at least one cell in an adjacent lower layer; and

wherein operations (a) to (f) are performed in one or more cells of the plurality of cells.

17. The computer readable medium of claim 16 wherein operations (a) to (f) are performed in each cell of the plurality of cells, and wherein the output of the at least one cell in each adjacent lower layer comprises an assumed state of a hidden markov process of the at least one cell.

18. The computer readable medium of claim 17 wherein the computer readable instructions further include instructions for performing operations comprising:

each cell in the top layer and the middle layers deriving from its predicted next state;

a predicted next state of a corresponding cell in its adjacent lower layer; and

each cell in the top layer and the middle layers transmitting to the corresponding cell the predicted next state of the corresponding cell.

19. The computer readable medium of claim 16 wherein said obtaining the predicted next state comprises:

obtaining from the state transition matrix the predicted next state by selecting from the state transition matrix a most likely state at a next step of the hidden markov process given said state.