Artificial neural network
According to an example aspect of the present invention, there is provided an apparatus comprising memory configured to store convolutional artificial neural network information comprising at least one filter definition, and at least one processing core configured to generate, from a preceding layer, a convolutional result of a succeeding layer of the artificial neural network in accordance with the at least one filter definition, and generate, from the convolutional result, an activation result of the succeeding layer by using an activation function, the activation function taking three arguments, the three arguments being derived from the convolutional result.
1. An apparatus comprising:
at least one processor; and
at least one memory including computer program code;
the at least one memory and the computer program code configured to, with the at least one processor, cause the apparatus at least to perform:
store convolutional artificial neural network information comprising at least one filter definition,
generate, from a preceding layer, a convolutional result of a succeeding layer of the artificial neural network in accordance with the at least one filter definition, and
generate, from the convolutional result, an activation result of the succeeding layer by using an activation function, the activation function taking three arguments, the three arguments being derived from the convolutional result.
2. The apparatus according to claim 1 , wherein one of the activation function arguments is a first convolutional result value, and the at least one processing core is configured to derive the other two of the activation function arguments from a neighbourhood of the first convolutional result value.
3. The apparatus according to claim 2 , wherein the neighbourhood of the first convolutional result value comprises eight other convolutional result values.
4. The apparatus according to claim 2 , wherein the other two of the activation function arguments comprise a value indicating a horizontal variation and a value indicating vertical variation.
5. The apparatus according to claim 1 , wherein the activation function produces, from the three arguments, two outputs.
6. The apparatus according to claim 5 , wherein the two outputs comprise an output relating to horizontal variation.
7. The apparatus according to claim 5 , wherein the two outputs comprise an output relating to vertical variation.
8. The apparatus according to claim 1 , wherein the last layer of the artificial neural network comprises a result layer.
9. The apparatus according to claim 1 , wherein the artificial neural network is configured to detect one or more objects.
10. The apparatus according to claim 1 , wherein the artificial neural network is configured to use an input data item which comprises an image.
11. A method comprising:
storing convolutional artificial neural network information comprising at least one filter definition,
generating, from a preceding layer, a convolutional result of a succeeding layer of the artificial neural network in accordance with the at least one filter definition, and
generating, from the convolutional result, an activation result of the succeeding layer by using an activation function, the activation function taking three arguments, the three arguments being derived from the convolutional result.
12. The method according to claim 11 , wherein one of the activation function arguments is a first convolutional result value, and the other two of the activation function arguments are derived from a neighbourhood of the first convolutional result value.
13. The method according to claim 12 , wherein the neighbourhood of the first convolutional result value comprises eight other convolutional result values.
14. The method according to claim 12 , wherein the other two of the activation function arguments comprise a value indicating a horizontal variation and a value indicating vertical variation.
15. The method according to claim 11 , wherein the activation function produces, from the three arguments, two outputs.
16. The method according to claim 15 , wherein the two outputs comprise an output relating to horizontal variation.
17. The method according to claim 15 , wherein the two outputs comprise an output relating to vertical variation.
18. The method according to claim 11 , wherein the last layer of the artificial neural network comprises a result layer.
19. The method according to claim 11 , wherein the artificial neural network is configured to detect one or more objects.
20. A non-transitory computer readable medium having stored thereon a set of computer readable instructions that, when executed by at least one processor, cause an apparatus to at least:
store convolutional artificial neural network information comprising at least one filter definition;
generate, from a preceding layer, a convolutional result of a succeeding layer of the artificial neural network in accordance with the at least one filter definition, and
generate, from the convolutional result, an activation result of the succeeding layer by using an activation function, the activation function taking three arguments, the three arguments being derived from the convolutional result.
21. The apparatus according to claim 1 ,
wherein the apparatus comprises at least one camera; and
wherein the convolutional artificial neural network is trained to detect one or more objects from an input data item which comprises a still image, a video image frame, or a short video clip.
22. The apparatus according to claim 21 , wherein the apparatus is a vehicle.
23. The apparatus according to claim 21 , wherein the apparatus is a smartphone.
24. The method according to claim 11 , wherein the convolutional artificial neural network is trained to detect one or more objects from a camera input data item which comprises a still image, a video image frame, or a short video clip.
25. The method according to claim 24 , wherein the method is equipped in a vehicle.
26. The method according to claim 24 , wherein the method is equipped in a smartphone.