Information processing apparatus using multi-layer neural network and method therefor

1. An information processing apparatus comprising:

one or more processors; and

at least one memory coupled to the one or more processors, the memory having stored thereon instructions which, when executed by the one or more processors, cause the apparatus to:

learn a plurality of multi-layer neural networks configured to carry out a plurality of tasks;

generate a shared layer candidate at a predetermined layer between or among the plurality of multi-layer neural networks by carrying out clustering to divide learning parameters at the predetermined layer into a plurality of clusters and selecting a representative learning parameter from the plurality of clusters;

relearn the plurality of multi-layer neural networks in a structure using the shared layer candidate; and

determine whether the shared layer candidate is to be shared at the predetermined layer with respect to each of the plurality of tasks based on an evaluation of the relearning.

2. The information processing apparatus according to claim 1 , wherein the predetermined layer is at least one layer in the multi-layer neural networks, and

wherein the information processing apparatus further determines a final structure of the plurality of multi-layer neural networks based on a result of the determination with respect to the shared layer candidate generated for each of the at least one layer.

3. The information processing apparatus according to claim 2 , further output the structure of the plurality of multi-layer neural networks that is determined.

4. The information processing apparatus according to claim 2 , further relearn the plurality of multi-layer neural networks having the structure determined.

5. The information processing apparatus according to claim 4 , further relearn in the structure of the plurality of multi-layer neural networks using the shared layer candidate, in a case where the predetermined layer includes a plurality of layers, determine every time that the shared layer candidate is to be shared with respect to one of the predetermined layers.

6. The information processing apparatus according to claim 5 , further comprising:

evaluate a performance of each of the tasks relearned; and

output the evaluation.

7. The information processing apparatus according to claim 1 , wherein the shared layer candidate by carrying out the clustering to divide the learning parameters including learning parameters at layers preceding and subsequent to the predetermined layer into the plurality of clusters and selecting the representative learning parameter from this plurality of clusters is generated.

8. The information processing apparatus according to claim 1 , wherein the generation carries out the clustering based on a similarity, and the similarity is any of an image similarity with the learning parameter assumed to be an image filter, a similarity in an output result when the image filter is applied to a predetermined image, a similarity in a result when predetermined non-linear processing is applied to the similarity in the output result when the image filter is applied to the predetermined image, and a similarity in a result when predetermined pooling processing is performed on the similarity in the result when the predetermined non-linear processing is applied.

9. The information processing apparatus according to claim 1 , wherein the shared layer candidate by randomly sampling a learning parameter at the predetermined layer is generated.

10. The information processing apparatus according to claim 1 , wherein the shared layer candidate by compressing data about a learning parameter at the predetermined layer is generated.

11. The information processing apparatus according to claim 1 , wherein the evaluation of the relearning is whether accuracy of the multi-layer neural networks in a case where the shared layer candidate is used falls within an allowable range.

12. The information processing apparatus according to claim 1 , wherein, the shared layer candidate with respect to a layer next to a last shared layer candidate in the predetermined layer range is generated in a case where the plurality of multi-layer neural networks already includes the shared layer candidate in a predetermined layer range.

13. A recognition processing apparatus configured to perform recognition processing by the plurality of multi-layer neural networks acquired by the information processing apparatus according to claim 1 .

14. An information processing method comprising:

learning a plurality of multi-layer neural networks configured to carry out a plurality of tasks;

generating a shared layer candidate at a predetermined layer between or among the plurality of multi-layer neural networks by carrying out clustering to divide learning parameters at the predetermined layer into a plurality of clusters and selecting a representative learning parameter from the plurality of clusters; and

relearning the plurality of multi-layer neural networks in a structure using the shared layer candidate, and determining whether the shared layer candidate is to be shared at the predetermined layer with respect to each of the plurality of tasks based on an evaluation of this relearning.

15. A non-transitory computer-readable storage medium storing a program for causing a computer to:

learn a plurality of multi-layer neural networks configured to carry out a plurality of tasks;

a generation unit configured to generate a shared layer candidate at a predetermined layer between or among the plurality of multi-layer neural networks by carrying out clustering to divide learning parameters at the predetermined layer into a plurality of clusters and selecting a representative learning parameter from the plurality of clusters;

relearn the plurality of multi-layer neural networks in a structure using the shared layer candidate; and

determine whether the shared layer candidate is to be shared at the predetermined layer with respect to each of the plurality of tasks based on an evaluation of the relearning.