PERSONALIZED SELECTION OF INFERENCE MODELS

1 . A method for personalized selection of inference models based on usage of other inference models, the method comprising:

receiving an indication of a usage of a first inference model by a first personalized device;

receiving an indication of a usage of a second inference model by a second personalized device;

selecting a first additional inference model based on the first inference model;

selecting a second additional inference model based on the second inference model, wherein the second additional inference model differs from the first additional inference model;

in response to the received indication of usage of the first inference model by the first personalized device, providing to the first personalized device information related to the selected first additional inference model; and

in response to the received indication of usage of the second inference model by the second personalized device, providing to the second personalized device information related to the selected second additional inference model.

2 . The method of claim 1 , wherein the selection of the first additional inference model is based on a compatibility of an output of the first inference model and an input of the first additional inference model, and wherein the selection of the second additional inference model is based on a compatibility of an output of the second inference model and an input of the second additional inference model.

3 . The method of claim 1 , wherein the selection of the first additional inference model is based on a compatibility of an input of the first inference model and an output of the first additional inference model, and wherein the selection of the second additional inference model is based on a compatibility of an input of the second inference model and an output of the second additional inference model.

4 . The method of claim 1 , wherein the selection of the first additional inference model is based on a compatibility of an input of the first inference model and an input of the first additional inference model, and wherein the selection of the second additional inference model is based on a compatibility of an input of the second inference model and an input of the second additional inference model.

5 . The method of claim 1 , wherein the selection of the first additional inference model is based on a compatibility of an output of the first inference model and an output of the first additional inference model, and wherein the selection of the second additional inference model is based on a compatibility of an output of the second inference model and an output of the second additional inference model.

6 . The method of claim 1 , wherein the received indication of the usage of the first inference model by the first personalized device comprises information related to outputs produced by the first inference model when used by the first personalized device, the received indication of the usage of the second inference model by the second personalized device comprises information related to outputs produced by the second inference model when used by the second personalized device, the selection of the first additional inference model is based on the information related to the outputs produced by the first inference model when used by the first personalized device, and the selection of the second additional inference model is based on the information related to the outputs produced by the second inference model when used by the second personalized device.

7 . The method of claim 1 , wherein the received indication of the usage of the first inference model by the first personalized device comprises information related to classification errors of the first inference model when used by the first personalized device, the received indication of the usage of the second inference model by the second personalized device comprises information related to classification errors of the second inference model when used by the second personalized device, the selection of the first additional inference model is based on the information related to the classification errors of the first inference model when used by the first personalized device, and the selection of the second additional inference model is based on the information related to the classification errors of the second inference model when used by the second personalized device.

8 . The method of claim 1 , wherein the received indication of the usage of the first inference model by the first personalized device comprises information related to confidence levels associated with outputs produced by the first inference model when used by the first personalized device, the received indication of the usage of the second inference model by the second personalized device comprises information related to confidence levels associated with outputs produced by the second inference model when used by the second personalized device, the selection of the first additional inference model is based on the information related to the confidence levels associated with the outputs produced by the first inference model when used by the first personalized device, and the selection of the second additional inference model is based on the information related to the confidence levels associated with the outputs produced by the second inference model when used by the second personalized device.

9 . The method of claim 1 , wherein the received indication of the usage of the first inference model by the first personalized device comprises information related to usage time of the first inference model by the first personalized device, the received indication of the usage of the second inference model by the second personalized device comprises information related to usage time of the second inference model by the second personalized device, the selection of the first additional inference model is based on the information related to the usage time of the first inference model by the first personalized device, and the selection of the second additional inference model is based on the information related to the usage time of the second inference model by the second personalized device.

10 . The method of claim 1 , wherein the received indication of the usage of the first inference model by the first personalized device comprises information related to locations of usage of the first inference model by the first personalized device, the received indication of the usage of the second inference model by the second personalized device comprises information related to location of usage of the second inference model by the second personalized device, the selection of the first additional inference model is based on the information related to the location of usage of the first inference model by the first personalized device, and the selection of the second additional inference model is based on the information related to the location of usage of the second inference model by the second personalized device.

11 . The method of claim 1 , wherein the received indication of the usage of the first inference model by the first personalized device comprises information related to inputs provided to the first inference model by the first personalized device, the received indication of the usage of the second inference model by the second personalized device comprises information related to inputs provided to the second inference model by the second personalized device, the selection of the first additional inference model is based on the information related to the inputs provided to the first inference model by the first personalized device, and the selection of the second additional inference model is based on the information related to the inputs provided to the second inference model by the second personalized device.

12 . The method of claim 1 , further comprising suggesting the first additional inference model selected based on the first inference model to a user of the first personalized device.

13 . The method of claim 1 , further comprising causing the first personalized device to use a cascade of the first inference model and the first additional inference model selected based on the first inference model.

14 . The method of claim 1 , further comprising causing the first personalized device to use an aggregate of outputs of the first inference model corresponding to an input and of outputs of the first additional inference model corresponding to the input to determine an overall output corresponding to the input.

15 . The method of claim 1 , further comprising causing the first personalized device to:

provide a first input to the first inference model to determine a first output;

provide a second input to the first additional inference model selected based on the first inference model to determine a second output, where the second input differs from the first input; and

combine the first output and the second output to determine an overall output corresponding to a combination of the first input and the second input.

16 . The method of claim 1 , wherein the received indication of the usage of the first inference model by the first personalized device comprises an indication that the first personalized device uses the first inference model to handle at least a first type of input and a second type of input, and wherein the method further comprising, in response to the received indication of usage of the first inference model by the first personalized device, causing the first personalized device to:

use the first inference model to handle at least the first type of input; and

use the first additional inference model selected based on the first inference model to handle at least the second type of input.

17 . The method of claim 1 , wherein the received indication of the usage of the first inference model by the first personalized device comprises an indication that the first personalized device uses the first inference model to handle at least inputs received at a first time window and inputs received at a second time window, and wherein the method further comprising, in response to the received indication of usage of the first inference model by the first personalized device, causing the first personalized device to:

use the first inference model to handle at least inputs received at the first time window; and

use the first additional inference model selected based on the first inference model to handle at least inputs received at the second time window.

18 . The method of claim 1 , wherein the received indication of the usage of the first inference model by the first personalized device comprises an indication that the first personalized device uses the first inference model to handle at least inputs associated with a first geographical area and inputs associated with a second geographical area, and wherein the method further comprising, in response to the received indication of usage of the first inference model by the first personalized device, causing the first personalized device to:

use the first inference model to handle at least inputs associated with the first geographical area; and

use the first additional inference model selected based on the first inference model to handle at least inputs associated with the second geographical area.

19 . A system for personalized selection of inference models based on usage of other inference models, the system comprising:

at least one processor configured to:

receive an indication of a usage of a first inference model by a first personalized device;

receive an indication of a usage of a second inference model by a second personalized device;

select a first additional inference model based on the first inference model;

select a second additional inference model based on the second inference model, wherein the second additional inference model differs from the first additional inference model;

in response to the received indication of usage of the first inference model by the first personalized device, provide to the first personalized device information related to the selected first additional inference model; and

in response to the received indication of usage of the second inference model by the second personalized device, provide to the second personalized device information related to the selected second additional inference model.

20 . A non-transitory computer readable medium storing data and computer implementable instructions for carrying out a method for personalized selection of inference models based on usage of other inference models, the method comprising:

receiving an indication of a usage of a first inference model by a first personalized device;

receiving an indication of a usage of a second inference model by a second personalized device;

selecting a first additional inference model based on the first inference model;

selecting a second additional inference model based on the second inference model, wherein the second additional inference model differs from the first additional inference model;

in response to the received indication of usage of the first inference model by the first personalized device, providing to the first personalized device information related to the selected first additional inference model; and

in response to the received indication of usage of the second inference model by the second personalized device, providing to the second personalized device information related to the selected second additional inference model.