SELECTIVE UPDATE OF INFERENCE MODELS
Systems and methods for selective update of inference models are provided. For example, a first inference model and a second inference model may be obtained, and the first inference model and the second inference model are based on a first version of a set of training examples. Further, in some examples, an updated version of the set of training examples may be obtained. Further, in some examples, an updated version of the first inference model may be generated based on the updated version of the set of training examples. Further, in some examples, the updated version of the first inference model may be evaluated, and the evaluation result may be used to determine whether to update the second inference model.
1 . A method for selective update of inference models, the method comprising:
obtaining a first inference model and a second inference model, the first inference model and the second inference model are based on a first version of a set of training examples;
obtaining an updated version of the set of training examples;
based on the updated version of the set of training examples, generating an updated version of the first inference model;
evaluating the updated version of the first inference model to obtain an evaluation result; and
using the evaluation result to determine whether to update the second inference model.
2 . The method of claim 1 , wherein the first inference model is a result of training a machine learning algorithm using a first training set, the second inference model is a result of training the machine learning algorithm using a second training set, the second training set differs from the first training set, and both the first training set and the second training set comprises the set of training examples.
3 . The method of claim 1 , wherein the first inference model is a result of training a first machine learning algorithm using the set of training examples, the second inference model is a result of training a second machine learning algorithm using the set of training examples, the updated version of the first inference model is a result of training the first machine learning algorithm using the updated version of the set of training examples, and wherein updating the second inference model comprises training the second machine learning algorithm using the updated version of the set of training examples.
4 . The method of claim 1 , wherein the first inference model is a result of training a machine learning algorithm using the set of training examples and a first hyper-parameter of the machine learning algorithm, the second inference model is a result of training the machine learning algorithm using the set of training examples and a second hyper-parameter of the machine learning algorithm, the updated version of the first inference model is a result of training the machine learning algorithm using the updated version of the set of training examples and the first hyper-parameter of the machine learning algorithm, and wherein updating the second inference model comprises training the machine learning algorithm using the updated version of the set of training examples and the second hyper-parameter of the machine learning algorithm.
5 . The method of claim 1 , wherein the first inference model is used by a first personalized device, the second inference model is used by a second personalized device, and further comprising:
in response to a first evaluation result, causing the second personalized device to use an updated version of the second inference model; and
in response to a second evaluation result, forgoing causing the second personalized device to use the updated version of the second inference model.
6 . The method of claim 1 , wherein the first inference model is used by a first personalized device, the second inference model is used by a second personalized device, and further comprising:
causing the first personalized device to use the updated version of the first inference model;
receiving results of the usage of the updated version of the first inference model by the first personalized device; and
using the received results of the usage of the updated version of the first inference model by the first personalized device to evaluate the updated version of the first inference model and obtain the evaluation result.
7 . The method of claim 1 , wherein the first inference model and the second inference model are used by a first personalized device, and further comprising causing a second personalized device to use the updated version of the first inference model and an updated version of the second inference model.
8 . The method of claim 1 , further comprising:
using the first inference model to analyze a plurality of inputs and produce a first plurality of outputs corresponding to the inputs;
using the updated version of the first inference model to analyze the plurality of inputs and produce a second plurality of outputs corresponding to the inputs; and
using the first plurality of outputs produced using the first inference model and the second plurality of outputs produced using the updated version of the first inference model to evaluate the updated version of the first inference model and obtain the evaluation result.
9 . The method of claim 1 , further comprising:
using the updated version of the first inference model to analyze a plurality of inputs and produce a plurality of outputs corresponding to the inputs;
providing information based on the plurality of outputs to at least one user;
receiving one or more feedbacks related to the information based on the plurality of outputs from the at least one user; and
using the one or more feedbacks related to the information based on the plurality of outputs to evaluate the updated version of the first inference model and obtain the evaluation result.
10 . The method of claim 1 , wherein the first inference model is configured to analyze visual data captured from an environment, and the second inference model is configured to analyze audio data captured in conjunction with the visual data.
11 . The method of claim 1 , wherein the first inference model is configured to analyze audio data captured from an environment, and the second inference model is configured to analyze visual data captured in conjunction with the audio data.
12 . The method of claim 1 , wherein the first inference model is configured to analyze audio data captured from an environment, and the second inference model is configured to analyze textual information associated with the audio data.
13 . The method of claim 1 , wherein the first inference model is configured to analyze image data captured from an environment, and the second inference model is configured to analyze textual information associated with the image data.
14 . The method of claim 1 , wherein the first inference model is configured to detect a first type of items in image data captured from an environment, and the second inference model is configured to determine properties of the detected items by analyzing the image data.
15 . The method of claim 1 , wherein the first version of the set of training examples includes at least one example not included in the updated version of the set of training examples, and further comprising using the at least one example not included in the updated version of the set of training examples to generate the updated version of the first inference model.
16 . The method of claim 1 , wherein the updated version of the set of training examples includes at least one example not included in the first version of the set of training examples, and further comprising using the at least one example not included in the first version of the set of training examples to generate the updated version of the first inference model.
17 . A system for selective update of inference models, the system comprising:
at least one processor configured to:
obtain a first inference model and a second inference model, the first inference model and the second inference model are based on a first version of a set of training examples;
obtain an updated version of the set of training examples;
based on the updated version of the set of training examples, generate an updated version of the first inference model;
evaluate the updated version of the first inference model to obtain an evaluation result; and
use the evaluation result to determine whether to update the second inference model.
18 . The system of claim 17 , wherein the first inference model is configured to analyze visual data captured from an environment, and the second inference model is configured to analyze audio data captured in conjunction with the visual data.
19 . The system of claim 17 , wherein the first inference model is configured to analyze audio data captured from an environment, and the second inference model is configured to analyze visual data captured in conjunction with the audio data.
20 . A non-transitory computer readable medium storing data and computer implementable instructions for carrying out a method for selective update of inference models, the method comprising:
obtaining a first inference model and a second inference model, the first inference model and the second inference model are based on a first version of a set of training examples;
obtaining an updated version of the set of training examples;
based on the updated version of the set of training examples, generating an updated version of the first inference model;
evaluating the updated version of the first inference model to obtain an evaluation result; and
using the evaluation result to determine whether to update the second inference model.