Method of generalized balanced few-shot learning using only novel data without old data
AssigneeInha University Research and Business Foundation
InventorsDongwan Choi, Seong Woong Kim
Life of this patent
Estimating the term…
Abstract
A method and system of generalized balanced few-shot learning using only novel data without old data is proposed. The method of generalized balanced few-shot learning using only novel data without old data proposed in the present disclosure includes pre-training stage for training a feature extractor and a classifier of a training model with base data through a pre-training unit and fine-tuning stage for freezing the feature extractor through a fine-tuning unit, training a joint linear classifier capable of inferring base classes and novel classes, and performing weight normalization to achieve zero-mean and balanced variance.
Claims6
A method of few-shot learning, comprising:
- training a feature extractor and a classifier of a training model with base data through a pre-training unit in a pre-training stage;
- and
- freezing the feature extractor through a fine-tuning unit, training a joint linear classifier capable of inferring base classes and novel classes, and performing a weight normalization to achieve a zero-mean and balanced variance of the joint linear classifier in a fine-tuning stage, wherein the fine-tunning stage further comprises:
- performing normalization during a training process to keep a weight mean of a novel classifier at zero;
- adjusting weights of a base classifier to match a ratio of standard deviations after completing the fine-tuning stage, wherein the ratio of standard deviations includes the standard deviation of the novel classes and the base classes;
- and
- optimizing decision boundaries of the novel classes by using class-wise learnable parameters.
A system of few-shot learning, comprising:
- a pre-training unit for training a feature extractor and a classifier of a training model with base data;
- and
- a fine-tuning unit for freezing the feature extractor, training a joint linear classifier capable of inferring base classes and novel classes, and performing weight normalization to achieve zero-mean and balanced variance of the joint linear classifier, wherein the fine-tuning unit includes a novel classifier and a base classifier, performs normalization during a training process to keep a weight mean of the novel classifier at zero, after completing the fine-tuning, adjusts the weights of the base classifier to match a ratio of standard deviations, wherein the ratio of standard deviations include the standard deviation of the novel classes and the base classes, and optimizes decision boundaries of the novel classes by using class-wise learnable parameters.
Dependencies are inferred from the claim text.