Machine learning system, method, and program product for point of sale systems

1. A method comprising steps of:

scanning, with a scanner system, at least one identification means associated with an object to acquire an object identifier determined from said scanned at least one identification means;

obtaining a classification ID of the object stored in a database based on said object identifier;

capturing, with an imaging device, an image of the object;

communicating said captured image of said object to a training system;

communicating said classification ID to said training system;

training a neural network system with said captured image and first classification ID to produce at least one of, a synaptic weight and a neural weight;

receiving, with a prediction module, said at least one of, a synaptic weight and a neural weight from said training system;

predicting, with said prediction module, a second classification ID of said captured image, said predicting using at least said synaptic weight;

comparing said first classification ID and said second classification ID;

determining an error rate based on said comparing step; and

classifying objects using said prediction module if said error rate is at a predetermined acceptable level.

2. The method as recited in claim 1 , further comprising a step of retrieving said classification ID from said database using said identifier obtained from the at least one identification means.

3. The method as recited in claim 1 , further comprising a step of accepting a manual entry for said classification ID.

4. The method as recited in claim 1 , further comprising a step of storing said synaptic weight for said predicting step if said error rate is at a predetermined acceptable level.

5. The method as recited in claim 4 , further comprising a step of determining a classifying ID for an unknown object using at least said stored synaptic weight.

6. The method as recited in claim 4 , further comprising steps of capturing an image of an unknown object with a mobile device and predicting, with said prediction module, a classification ID of said captured image, using at least said stored synaptic weight.

7. The method as recited in claim 1 , in which said neural network system comprises at least one convolutional layer that is configured to compute a forward pass and a backpropagation pass, and in which said convolutional layer comprises at least a trainable filter and at least one trainable bias per feature map from a plurality of pooled feature maps.

8. The method as recited in claim 7 , in which said neural network system further comprises at least one pooling layer that is configured to achieve a spatial invariance by reducing a resolution of said feature map, wherein each pooled feature map from said plurality of pooled feature maps correspond to one feature map of a previous map layer.

9. A non-transitory computer-readable storage medium with an executable program stored thereon, wherein the program instructs one or more processors to perform the following steps:

scanning, with a mobile device, at least one identification means associated with an object to acquire an object identifier determined from said scanned at least one identification means;

obtaining a first classification ID stored in a database based on said object identifier;

capturing an image of the object;

communicating said first classification ID and the captured image to a training system, the training system being configured for at least training a neural network system with said first classification ID and said captured image to produce at least one synaptic weight;

receiving, with a prediction module, said synaptic weight from the training system; and

predicting, with said prediction module, a second classification ID of said captured image, said predicting using at least the synaptic weight;

comparing said first classification ID and said second classification ID;

determining an error rate based on said comparing step;

storing said synaptic weight if said error rate is at a predetermined acceptable level; and

classifying objects using said prediction module based on said stored synaptic weight.

10. The non-transitory computer-readable storage medium of claim 9 , wherein the program further comprises a step of capturing an image of an unknown object with a mobile device.

11. The non-transitory computer-readable storage medium of claim 10 , wherein the program further comprises a step of communicating said captured image of an unknown object to said prediction module.

12. The non-transitory computer-readable storage medium of claim 11 , wherein the program further comprises a step of predicting, with said prediction module, a classification ID of said captured image of an unknown object, using at least said stored synaptic weight.

13. The non-transitory computer-readable storage medium of claim 9 , wherein the program further comprises steps of:

retrieving the first classification ID from said database using said object identifier obtained from the at least one identification means;

accepting a manual entry for the first classification ID; in which said comparing step comprises comparing a predicted identification using said synaptic weight to the object identifier to produce an error rate; and

predicting a classification of a captured image of unknown objects with said prediction module within a point of sale system wherein said captured image identify an item for sale.