Method and system for dynamic latent vector allocation

1 . A method, implemented on a machine having at least one processor, storage, and a communication platform capable of connecting to a network for performance based dynamic vector construction via machine-training, the method comprising:

machine-training, by a training engine of a dynamic vector allocator implemented by a processor based on a target vector, a machine-trained model, wherein the target vector has a plurality of feature entries, wherein each of the plurality of feature entries is mapped from at least one attribute from at least first and second source vectors, and wherein a number of feature entries in the target vector associated with attributes from the first source vector is same as that in the target vector associated with attributes from the second source vector;

based on a timing model, activating, by a trigger of the dynamic vector allocator implemented by the processor, a feature vector modifier of the dynamic vector allocator to adjust, based on informative values of the first and second source vectors, the number of feature entries in the target vector corresponding to attributes from the first source vector and the number of feature entries in the target vector corresponding to attributes from the second source vector, by:

identifying, by the feature vector modifier of the dynamic vector allocator implemented by the processor, one of the feature entries in the target vector, in accordance with a first criterion via machine learning based on the machine-trained model, for replacing the corresponding at least one attribute from the first source vector;

determining, by the feature vector modifier of the dynamic vector allocator, at least one alternative attribute from the second source vector based on a second criterion, wherein the at least one alternative attribute is to be mapped to the feature entry of the target vector; and

updating, by a reallocator of the dynamic vector allocator implemented by the processor, the target vector by populating the feature entry of the target vector based on the at least one alternative attribute, wherein the updating comprises changing the numbers of feature entries in the target vector associated with attributes respectively from the first source vector and from the second source vector so that a number of feature entries in the updated target vector associated with attributes from the first source vector is different from that associated with attributes from the second source vector;

feeding back, by the reallocator of the dynamic vector allocator, the updated target vector to the training engine of the dynamic vector allocator; and

further machine-training, by the training engine of the dynamic vector allocator using the fed-back updated target vector, the machine-trained model to focus more on data associated with one of the first and second source vectors that holds more informative value than on data associated with the other one of the first and second source vectors that holds less informative value, thereby reducing a training time of the training engine compared to user feature vector training engines that consider the first and second source vectors to be equally important.

2 . The method of claim 1 , wherein the first source vector has less informative value than the second source vector, and

the number of feature entries in the updated target vector associated with attributes from the first source vector is less than that associated with attributes from the second source vector.

3 . The method of claim 1 , wherein the machine learning is related to determining effectiveness of the plurality of feature entries in contributing to the machine learning.

4 . The method of claim 1 , wherein the first criterion corresponds to selecting the feature entry in the target vector that has a minimum loss value.

5 . The method of claim 1 , wherein the second criterion corresponds to selecting, from a plurality of alternative attributes, the at least one alternative attribute which incurs a minimum loss value upon being populated to the feature entry of the target vector.

6 . The method of claim 1 , wherein the first source vector partially overlaps with the second source vector.

7 . The method of claim 1 , wherein the feature entry of the target vector identified to be mapped using the at least one alternative attribute is assessed as not being effective in contributing to the machine learning.

8 . A non-transitory machine-readable medium having information recorded thereon for performance based dynamic vector construction via machine-training, wherein the information, when read by a machine, causes the machine to perform the steps of:

machine-training, by a training engine of a dynamic vector allocator implemented by a processor based on a target vector, a machine-trained model, wherein the target vector has a plurality of feature entries, wherein each of the plurality of feature entries is mapped from at least one attribute from at least first and second source vectors, and wherein a number of feature entries in the target vector associated with attributes from the first source vector is same as that in the target vector associated with attributes from the second source vector;

based on a timing model, activating, by a trigger of the dynamic vector allocator implemented by the processor, a feature vector modifier of the dynamic vector allocator to adjust, based on informative values of the first and second source vectors, the number of feature entries in the target vector corresponding to attributes from the first source vector and the number of feature entries in the target vector corresponding to attributes from the second source vector, by:

identifying, by the feature vector modifier of the dynamic vector allocator implemented by the processor, one of the feature entries in the target vector, in accordance with a first criterion via machine learning based on the machine-trained model, for replacing the corresponding at least one attribute from the first source vector;

determining, by the feature vector modifier of the dynamic vector allocator, at least one alternative attribute from the second source vector based on a second criterion, wherein the at least one alternative attribute is to be mapped to the feature entry of the target vector; and

updating, by a reallocator of the dynamic vector allocator implemented by the processor, the target vector by populating the feature entry of the target vector based on the at least one alternative attribute, wherein the updating comprises changing the numbers of feature entries in the target vector associated with attributes respectively from the first source vector and from the second source vector so that a number of feature entries in the updated target vector associated with attributes from the first source vector is different from that associated with attributes from the second source vector;

feeding back, by the reallocator of the dynamic vector allocator, the updated target vector to the training engine of the dynamic vector allocator; and

further machine-training, by the training engine of the dynamic vector allocator using the fed-back updated target vector, the machine-trained model to focus more on data associated with one of the first and second source vectors than on data associated with the other one of the first and second source vectors that holds less informative value, thereby reducing a training time of the training engine compared to user feature vector training engines that consider the first and second source vectors to be equally important.

9 . The medium of claim 8 , wherein the first source vector has less informative value than the second source vector, and

the number of feature entries in the updated target vector associated with attributes from the first source vector is less than that associated with attributes from the second source vector.

10 . The medium of claim 8 , wherein the machine learning is related to determining effectiveness of the plurality of feature entries in contributing to the machine learning.

11 . The medium of claim 8 , wherein the first criterion corresponds to selecting the feature entry in the target vector that has a minimum loss value.

12 . The medium of claim 8 , wherein the second criterion corresponds to selecting, from a plurality of alternative attributes, the at least one alternative attribute which incurs a minimum loss value upon being populated to the feature entry of the target vector.

13 . The medium of claim 8 , wherein the first source vector partially overlaps with the second source vector.

14 . The medium of claim 8 , wherein the feature entry of the target vector identified to be mapped using the at least one alternative attribute is assessed as not being effective in contributing to the machine learning.

15 . A system for performance based dynamic vector construction via machine-training, the system comprising: a processor, and an engine implemented by the processor and configured to:

machine-train, by a training engine, based on a target vector, a machine-trained model, wherein the target vector has a plurality of feature entries, wherein the target vector has a plurality of feature entries, wherein each of the plurality of feature entries is mapped from at least one attribute from at least first and second source vectors, and wherein a number of feature entries in the target vector associated with attributes from the first source vector is same as that in the target vector associated with attributes from the second source vector;

based on a timing model, activate, by a trigger, a feature vector modifier to adjust, based on informative values of the first and second source vectors, the number of feature entries in the target vector corresponding to attributes from the first source vector and the number of feature entries in the target vector corresponding to attributes from the second source vector, by:

identifying one of the feature entries in the target vector, in accordance with a first criterion via the machine learning based on the machine-trained model, for replacing the corresponding at least one attribute from the first source vector;

determining at least one alternative attribute from the second source vector based on a second criterion, wherein the at least one alternative attribute is to be mapped to the feature entry of the target vector; and

updating the target vector by populating the feature entry of the target vector based on the at least one alternative attribute, wherein the updating comprises changing the numbers of feature entries in the target vector associated with attributes respectively from the first source vector and from the second source vector so that a number of feature entries in the updated target vector associated with attributes from the first source vector is different from that associated with attributes from the second source vector;

feeding back the updated target vector to the training engine; and

further machine-training, by the training engine using the fed-back updated target vector, the machine-trained model to focus more on data associated with one of the first and second source vectors that holds more informative value than on data associated with the other one of the first and second source vectors that holds less informative value, thereby reducing a training time of the training engine compared to user feature vector training engines that consider the first and second source vectors to be equally important.

16 . The system of claim 15 , wherein the first source vector has less informative value than the original source vector, and

the number of feature entries in the updated target vector associated with attributes from the first source vector is less than that associated with attributes from the second source vector.

17 . The system of claim 15 , wherein the first criterion corresponds to selecting the feature entry in the target vector that has a minimum loss value.

18 . The system of claim 15 , wherein the second criterion corresponds to selecting, from a plurality of alternative attributes, the at least one alternative attribute which incurs a minimum loss value upon being populated to the feature entry of the target vector.

19 . The system of claim 15 , wherein the first source vector partially overlaps with the second source vector.

20 . The system of claim 15 , wherein the feature entry of the target vector identified to be mapped using the at least one alternative attribute is assessed as not being effective in contributing to the machine learning.