17527295. META FEW-SHOT CLASS INCREMENTAL LEARNING simplified abstract (HUAWEI TECHNOLOGIES CO., LTD.)
Contents
META FEW-SHOT CLASS INCREMENTAL LEARNING
Organization Name
Inventor(s)
Zhixiang Chi of North York (CA)
META FEW-SHOT CLASS INCREMENTAL LEARNING - A simplified explanation of the abstract
This abstract first appeared for US patent application 17527295 titled 'META FEW-SHOT CLASS INCREMENTAL LEARNING
Simplified Explanation
This patent application describes a method and system for meta few-shot class incremental learning. The method involves obtaining weight attention maps from a first network and using them to update the weights of a second network. The second network is a modulatory network. The method also includes generating feature attention maps of the second network based on the weight attention maps and a set of input images of at least one class. Additionally, feature maps of the first network are generated based on the input images, and the feature maps are updated using the feature attention maps of the second network.
- Obtaining weight attention maps from a first network
- Updating the weights of a second network using the weight attention maps
- Generating feature attention maps of the second network based on the weight attention maps and input images
- Generating feature maps of the first network based on the input images
- Updating the feature maps of the first network using the feature attention maps of the second network
Potential applications of this technology:
- Artificial intelligence and machine learning systems
- Computer vision and image recognition systems
- Incremental learning in various domains
Problems solved by this technology:
- Addressing the challenge of few-shot learning, where a model needs to learn new classes with limited training examples
- Enabling incremental learning, where a model can learn new classes without forgetting previously learned classes
Benefits of this technology:
- Improved performance in few-shot learning tasks
- Ability to learn new classes without forgetting previously learned classes
- Enhanced adaptability and flexibility of machine learning models
Original Abstract Submitted
This disclosure provides for methods and system for meta few-shot class incremental learning. According to an aspect a method is provided. The method includes obtaining at least one weight attention map of a first network and updating weights of a second network using the at least one weight attention map, where the second network is a modulatory network. The method further includes generating at least one feature attention map of the second network based on the at least one weight attention map of the first network and a set of input images of at least one class. The method further includes generating at least one feature map of the first network based on the set of input images of the at least one class, and updating the at least one feature map of the first network based on the feature attention map of the second network.