Recognition method and apparatus

1. A recognition method comprising:

determining a feature vector corresponding to an ensemble model from input data, based on the ensemble model; and

recognizing the input data based on the feature vector,

wherein the ensemble model comprises a first model and a second model with a structure that is the same as a structure of the first model,

wherein same input data among the input data is input to the first model and the second model,

wherein the determining of the feature vector comprises determining a feature vector output from a second layer based on a portion of a first operation result obtained based on first connection weights between a first layer and the second layer in the ensemble model, and a first feature vector output from the first layer based on the input data, and

wherein the portion of the first operation result comprises:

an operation result obtained based on first partial weights corresponding to the first model among the first connection weights, and a first partial vector corresponding to the first model in the first feature vector; and

an operation result obtained based on second partial weights corresponding to the second model among the first connection weights, and a second partial vector corresponding to the second model in the first feature vector.

2. The recognition method of claim 1 , wherein the determining of the feature vector further comprises using second connection weights of the second layer between a previous time and a current time, and a second feature vector output from the second layer at the previous time, to determine the feature vector output from the second layer.

3. The recognition method of claim 2 , wherein the determining of the feature vector further comprises determining the feature vector by using a portion of a second operation result obtained based on the second connection weights and the second feature vector.

4. The recognition method of claim 3 , wherein the portion of the second operation result comprises:

an operation result obtained based on third partial weights corresponding to the first model among the second connection weights and a third partial vector corresponding to the first model in the second feature vector; and

an operation result obtained based on fourth partial weights corresponding to the second model among the second connection weights and a fourth partial vector corresponding to the second model in the second feature vector.

5. The recognition method of claim 3 , wherein the portion of the second operation result comprises a principal diagonal element of a matrix determined as the second operation result.

6. The recognition method of claim 1 , wherein the first model and the second model are assigned to a same apparatus.

7. The recognition method of claim 1 , wherein the first model and the second model are implemented in parallel.

8. The recognition method of claim 1 , wherein the first model and the second model comprise a same number of layers, a same number of nodes and a same connection relationship between the layers.

9. The recognition method of claim 1 , wherein the recognizing of the input data comprises recognizing the input data based on an n-th element corresponding to the first model and an n-th element corresponding to the second model in the feature vector.

10. The recognition method of claim 1 , wherein the feature vector comprises a first partial feature vector corresponding to the first model and a second partial feature vector corresponding to the second model.

11. A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the method of claim 1 .

12. A recognition apparatus comprising:

a processor; and

a memory comprising at least one instruction executable by the processor,

wherein the processor is configured to, in response to the at least one instruction being executed by the processor,

determine a feature vector corresponding to an ensemble model from input data, based on the ensemble model, and

recognize the input data based on the feature vector, and

wherein the ensemble model comprises a first model and a second model with a structure that is the same as a structure of the first model,

wherein the processor is further configured to input same input data among the input data to the first model and the second model,

wherein the processor is further configured to determine the feature vector output from the second layer based on a portion of a first operation result obtained based on first connection weights between a first layer and a second layer in the ensemble model, and a first feature vector output from the first layer based on the input data, and

wherein the portion of the first operation result comprises:

an operation result obtained based on first partial weights corresponding to the first model among the first connection weights, and a first partial vector corresponding to the first model in the first feature vector; and

an operation result obtained based on second partial weights corresponding to the second model among the first connection weights, and a second partial vector corresponding to the second model in the first feature vector.

13. The recognition apparatus of claim 12 , wherein the processor is further configured to use second connection weights of the second layer between a previous time and a current time, and a second feature vector output from the second layer at the previous time, to determine the feature vector output from the second layer.

14. The recognition apparatus of claim 13 , wherein the processor is further configured to determine the feature vector by further using a portion of a second operation result obtained based on the second connection weights and the second feature vector.

15. The recognition apparatus of claim 14 , wherein the portion of the second operation result comprises a principal diagonal element of a matrix determined as the second operation result.

16. A method to recognize input data, comprising:

receiving the input data in an ensemble neural network comprising a first neural network and a second neural network implemented in parallel with the first neural network;

calculating, by a processor, in a first layer of the first neural network, a first partial feature vector;

calculating, by the processor, in a first layer of the second neural network, a second partial feature vector;

calculating, by the processor, a feature vector, by applying a weight matrix to the first partial feature vector and the second partial feature vector, wherein the weight matrix comprises connection weights between the first layer of the first neural network and a second layer of the first neural network, and connection weights between the first layer of the second neural network and a second layer of the second neural network; and

recognizing the input data based on the feature vector.

17. The method of claim 16 , wherein the applying of the weight matrix to the first partial feature vector and the second partial feature vector comprises a multiplying a matrix comprising the first partial feature vector and the second partial feature vector by the weight matrix.

18. The method of claim 17 , wherein the calculating of the feature vector further comprises applying an activation function to a result of the multiplying of the matrix comprising the first partial feature vector and the second partial feature vector by the weight matrix.

19. The method of claim 18 , wherein the activation function comprises a term based on a third partial feature vector output from the second layer of the first neural network at a previous time, and a fourth partial feature vector output from the second layer of the second neural network at a previous time.

20. The method of claim 16 , wherein the input data comprises one of an image, an audio signal, and a text.