# US 2022/0067443 A1 · System and method for model-agnostic meta-learner for noisy data with label errors

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2022/0067443 A1
- Document ID: US20220067443A1
- Status: Published application
- Kind: A1
- Application number: 17/009,977
- Filed: 2020-09-02
- Published: 2022-03-03
- Earliest claimed: 2020-09-02
- Assignees: Oath Inc.
- Applicants: Oath Inc.
- Inventors: Eliyar Asgarieh
- Web page: https://patentgrep.com/patent/US20220067443A1

## CPC

- G06F 18/2155: {characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling}
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06F 9/3016: {Decoding the operand specifier, e.g. specifier format}
- G06F 18/213: Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F 18/24: Classification techniques
- G06N 3/045: Combinations of networks
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/0895: Weakly supervised learning, e.g. semi-supervised or self-supervised learning
- G06N 3/09: Supervised learning
- G06N 3/0985: Hyperparameter optimisation; Meta-learning; Learning-to-learn
- G06N 20/00: Machine learning
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

The present teaching relates to method, system, medium, and implementations for machine learning. Machine learning is performed based on training data via a dual loop learning process that includes a first loop for data decoding learning and a second loop for label decoding learning. In the first loop, first parameters associated with decoding are updated to generate updated first parameters based on a first label, estimated via the decoding using the first parameters, and a second label, predicted via the label decoding using second parameters. In the second loop, the second parameters associated with the label decoding are updated to generate updated second parameters based on a third label, obtained via the decoding using the updated first parameters, and a ground truth label.

## Claims

### Claim 1

A method implemented on at least one machine including at least one processor, memory, and communication platform capable of connecting to a network for machine learning, the method comprising:
performing machine learning based on training data via a dual loop learning process having a first loop for data decoding learning and a second loop for label decoding learning, wherein
in the first loop, first parameters associated with the decoding are updated to generate updated first parameters based on a first label estimated via the decoding using the first parameters and a second label predicted via the label decoding using second parameters, and
in the second loop, the second parameters associated with the label decoding are updated to generate updated second parameters based on a third label obtained via the decoding using the updated first parameters and a ground truth label.

### Claim 2 (depends on 1)

The method of claim 1, wherein the first loop comprises:
receiving first input data from a source;
classifying, based on the first parameters, the first input data to have the first label;
receiving the second label predicted by the label decoding based on the input data and second parameters;
determining a source loss based on the first label and the second label; and
generating the updated first parameters based on the source loss.

### Claim 3 (depends on 2)

The method of claim 2, wherein the source provides training data with inconsistent, incorrect, and missing labels.

### Claim 4 (depends on 2)

The method of claim 2, wherein the first parameters associated with the decoding include encoder parameters and decoder parameters.

### Claim 5 (depends on 4)

The method of claim 4, wherein the step of classifying comprises:
encoding, based on the encoder parameters, the first input data to generate a feature vector; and
decoding, based on the decoder parameters, the first input data in accordance with the encoded feature vector to generate the first label.

### Claim 6 (depends on 1)

The method of claim 1, wherein the second loop comprises:
receiving second input data from a target, wherein the second input is associated with the ground truth label;
classifying the second input data to have the third label based on the updated first parameters;
determining a target loss based on the third label and the ground truth label; and
updating the second parameters based on the target loss.

### Claim 7 (depends on 6)

The method of claim 6, wherein the target provides training data with correctly labeled ground truth labels.

### Claim 8

Machine readable and non-transitory medium having information recorded thereon for machine learning, wherein the information, once read by the machine, causes the machine to perform:
performing machine learning based on training data via a dual loop learning process having a first loop for data decoding learning and a second loop for label decoding learning, wherein
in the first loop, first parameters associated with the decoding are updated to generate updated first parameters based on a first label estimated via the decoding using the first parameters and a second label predicted via the label decoding using second parameters, and
in the second loop, the second parameters associated with the label decoding are updated to generate updated second parameters based on a third label obtained via the decoding using the updated first parameters and a ground truth label.

### Claim 9 (depends on 8)

The medium of claim 8, wherein steps performed in the first loop comprises:
receiving first input data from a source;
classifying, based on the first parameters, the first input data to have the first label;
receiving the second label predicted by the label decoding based on the input data and second parameters;
determining a source loss based on the first label and the second label; and
generating the updated first parameters based on the source loss.

### Claim 10 (depends on 9)

The medium of claim 9, wherein the source provides training data with inconsistent, incorrect, and missing labels.

### Claim 11 (depends on 9)

The medium of claim 9, wherein the first parameters associated with the decoding include encoder parameters and decoder parameters.

### Claim 12 (depends on 11)

The medium of claim 11, wherein the step of classifying comprises:
encoding, based on the encoder parameters, the first input data to generate a feature vector; and
decoding, based on the decoder parameters, the first input data in accordance with the encoded feature vector to generate the first label.

### Claim 13 (depends on 8)

The medium of claim 8, wherein steps performed in the second loop comprises:
receiving second input data from a target, wherein the second input is associated with the ground truth label;
classifying the second input data to have the third label based on the updated first parameters;
determining a target loss based on the third label and the ground truth label; and
updating the second parameters based on the target loss.

### Claim 14 (depends on 13)

The medium of claim 13, wherein the target provides training data with correctly labeled ground truth labels.

### Claim 15

A system for machine learning, comprising:
a decoder configured for decoding training data; and
at least one label decoder configured for predicting a label based on training data, wherein
the decoder and the at least one label decoder collaborate in a dual loop learning process having a first loop for data decoding learning and a second loop for label decoding learning, wherein
in the first loop, first parameters associated with the decoding are updated to generate updated first parameters based on a first label classified by the decoder using the first parameters and a second label predicted by the at least one label decoder using second parameters, and
in the second loop, the second parameters associated with the at least one label decoders are updated to generate updated second parameters based on a third label classified by the decoder using the updated first parameters and a ground truth label.

### Claim 16 (depends on 15)

The system of claim 15, wherein the first parameters include encoder parameters used by an encoder to encode the training data and decoder parameters used by the decoder to classify the encoded training data.

### Claim 17 (depends on 16)

The system of claim 16, further comprising an encoder, wherein in the first loop:
the encoder is configured for
receiving a first input data from a source, and
encoding the first input data to generated encoded first input data based on the encoder parameters; and
the decoder is configured for classifying, based on the first parameters, the first input data to have the first label.

### Claim 18 (depends on 17)

The system of claim 17, further comprising
a source loss determiner configured for determining a source loss based on the first label and the second label; and
a decoder parameter updater configured for updating the first parameters based on the source loss to generate the updated first parameters.

### Claim 19 (depends on 17)

The system of claim 17, wherein each of the at least one label decoder comprises:
a label predictor configured for, in the first loop, predicting the second label for the first input data based on the second parameters;
a target label retriever configured for, in the second loop, retrieving a ground truth label for second input data from a target;
a target loss determiner configured for, in the second loop, computing a target loss based on the ground truth label and the third label generated by the decoder based on the second input data in accordance with the updated first parameters; and
a label prediction parameter updater configured for, in the second loop, updating the second parameters based on the target loss to generate the updated second parameters.

### Claim 20 (depends on 19)

The system of claim 19, wherein
the source provides training data with inconsistent, incorrect, and missing labels; and
the target provides training data with correctly labeled ground truth labels.
