Transfer learning and domain adaptation using distributable data models
A system for transfer learning and domain adaptation using distributable data models is provided, comprising a network-connected distributable model configured to serve instances of a plurality of distributable models; and a directed computation graph module configured to receive at least an instance of at least one of the distributable models from the network-connected computing system, create a second dataset from machine learning performed by a transfer engine, train the instance of the distributable model with the second dataset, and generate an update report based at least in part by updates to the instance of the distributable model.
1. A system for transfer learning and domain adaptation using distributable data models, comprising:
a distributable model source comprising a first plurality of programming instructions stored in a memory of, and operable on a processor of, a computing device, wherein the first plurality of programming instructions, when operating on the processor, cause the computing device to:
retrieve a distributable model;
instantiate a distributable model instance of the distributable model;
transfer the distributable model instance to an electronic device; and
receive a report from the electronic device, the report comprising updates to the distributable model instance made by the electronic device while training the distributable model instance on the electronic device; and
update the distributable model based on the report
a transfer engine comprising a second plurality of programming instructions stored in the memory of, and operable on the processor of, the computing device, wherein the second plurality of programming instructions, when operating on the processor, cause the computing device to:
receive a distributable model instance from the distributable model source and information about a desired adaptation of the distributable model instance;
apply one or more machine learning algorithms to the distributable model instance to adapt it to the environment or purpose; and
return the adapted distributable model instance to the distributable model source; and
a distributed computational graph engine comprising a third plurality of programming instructions stored in the memory of, and operable on the processor of, the computing device, wherein the third plurality of programming instructions, when operating on the processor, cause the computing device to:
receive the distributable model from the distributable model source;
receive one or more reports from distributable model instances which have been trained on one or more electronic devices;
create a bias-specific distributable model from the distributable model based on biases contained in the one or more reports;
return the bias-specific distributable model to the distributable model source; and
wherein the distributable model source incorporates the bias-specific distributable model as a subset of the distributable model.
2. The system of claim 1 , wherein the distributable model source further:
instantiates a plurality of distributable model instances;
transfers each distributable model instance to one of a plurality of electronic devices;
receives reports from the plurality of electronic devices; and
updates the distributable model based on the reports received.
3. The system of claim 1 , wherein the desired adaptation is adaptation to the computing environment of a type of electronic device.
4. The system of claim 1 , wherein the desired adaptation is adaptation of the distributable model instance for a particular purpose or use.
5. The system of claim 1 , wherein the desired adaptation is changing the domain of the distributable model instance.
6. A method for transfer learning and domain adaptation using distributable data models, comprising the steps of:
retrieving a distributable model;
instantiating a distributable model instance of the distributable model;
transferring the distributable model instance to an electronic device;
receiving a report from the electronic device, the report comprising updates to the distributable model instance made by the electronic device while training the distributable model instance on the electronic device;
updating the distributable model based on the report
receiving a distributable model instance from the distributable model source and information about a desired adaptation of the distributable model instance;
applying one or more machine learning algorithms to the distributable model instance to adapt it to the environment or purpose; and
returning the adapted distributable model instance to the distributable model source. receiving the distributable model from the distributable model source;
receiving one or more reports from distributable model instances which have been trained on one or more electronic devices;
creating a bias-specific distributable model from the distributable model based on biases contained in the one or more reports;
returning the bias-specific distributable model to the distributable model source; and
wherein the distributable model source incorporates the bias-specific distributable model as a subset of the distributable model.
7. The method of claim 6 , further comprising the steps of:
instantiating a plurality of distributable model instances;
transferring each distributable model instance to one of a plurality of electronic devices;
receiving reports from the plurality of electronic devices; and
updating the distributable model based on the reports received.
8. The method of claim 6 , wherein the desired adaptation is adaptation to the computing environment of a type of electronic device.
9. The method of claim 6 , wherein the desired adaptation is adaptation of the distributable model instance for a particular purpose or use.
10. The method of claim 6 , wherein the desired adaptation is changing the domain of the distributable model instance.