17984010. METHOD AND APPARATUS FOR CLASS INCREMENTAL LEARNING simplified abstract (SAMSUNG ELECTRONICS CO., LTD.)
METHOD AND APPARATUS FOR CLASS INCREMENTAL LEARNING
Organization Name
Inventor(s)
METHOD AND APPARATUS FOR CLASS INCREMENTAL LEARNING - A simplified explanation of the abstract
This abstract first appeared for US patent application 17984010 titled 'METHOD AND APPARATUS FOR CLASS INCREMENTAL LEARNING
Simplified Explanation
The present application describes a method for training a machine learning model using class incremental learning, and a computer-implemented method and apparatus for using the trained model.
- The method allows for updating semantic representations of old concepts by modeling drift of semantic representations.
- The method also enables updating feature representations of old concepts by modeling drift of feature representations.
Potential Applications
This technology has potential applications in various fields, including:
- Natural language processing: The method can be used to improve language understanding and generation models by updating semantic and feature representations of words and phrases.
- Computer vision: The method can enhance object recognition and image classification models by updating semantic and feature representations of objects and visual features.
- Recommender systems: The method can improve recommendation algorithms by updating semantic and feature representations of items and user preferences.
Problems Solved
This technology addresses the following problems:
- Concept drift: The method tackles the challenge of updating semantic and feature representations of old concepts as they change over time.
- Model performance degradation: By updating semantic and feature representations, the method helps prevent performance degradation of machine learning models when faced with new data or evolving concepts.
Benefits
The use of class incremental learning and modeling drift of semantic and feature representations offers several benefits:
- Improved accuracy: The method allows machine learning models to adapt to changes in concepts, leading to more accurate predictions and classifications.
- Efficient training: By selectively updating semantic and feature representations, the method reduces the computational resources required for retraining the entire model.
- Scalability: The method can handle a large number of classes and concepts, making it suitable for complex machine learning tasks.
Original Abstract Submitted
The present application generally relates to a method for training a machine learning, ML, model using class incremental learning, and to a computer-implemented method and apparatus for using the trained machine learning, ML, model. The method may learn how to update semantic representations of old concepts (classes) by modelling drift of semantic representations. The method may also learn how to update feature representations of old concepts (classes) by modelling drift of feature representations
