International business machines corporation (20240202515). CLASS-INCREMENTAL LEARNING OF A CLASSIFIER simplified abstract
Contents
CLASS-INCREMENTAL LEARNING OF A CLASSIFIER
Organization Name
international business machines corporation
Inventor(s)
Kumudu Geethan Karunaratne of Gattikon (CH)
Michael Andreas Hersche of Zurich (CH)
Giovanni Cherubini of Rueschlikon (CH)
Abu Sebastian of Adliswil (CH)
Abbas Rahimi of RUESCHLIKON (CH)
CLASS-INCREMENTAL LEARNING OF A CLASSIFIER - A simplified explanation of the abstract
This abstract first appeared for US patent application 20240202515 titled 'CLASS-INCREMENTAL LEARNING OF A CLASSIFIER
The present disclosure pertains to training a classifier with an explicit memory and a controller. The training process involves receiving multiple training datasets, adding output vectors to the memory, retraining the classifier, updating prototype vectors, and refreshing the memory.
- The classifier is trained iteratively with new datasets containing data samples of novel classes.
- Output vectors representing the novel classes are added to the explicit memory.
- The classifier is retrained using the new datasets and the original dataset by minimizing the distance between the output vectors and prototype vectors.
- Updated prototype vectors are determined based on the training datasets.
- The explicit memory is updated with the set of updated prototype vectors.
Potential Applications: - This technology can be used in image recognition systems to classify new objects. - It can be applied in natural language processing for identifying new language patterns. - This innovation can enhance the performance of recommendation systems by recognizing new user preferences.
Problems Solved: - Efficiently training classifiers with new data samples of novel classes. - Improving the accuracy of classifiers in recognizing unfamiliar patterns. - Enhancing the adaptability of machine learning models to evolving datasets.
Benefits: - Increased accuracy in classifying new data samples. - Improved performance of machine learning models in handling novel classes. - Enhanced adaptability and flexibility of classifiers in dynamic environments.
Commercial Applications: Title: Novel Classifier Training Technology for Enhanced Pattern Recognition This technology can be utilized in various industries such as healthcare for disease diagnosis, finance for fraud detection, and e-commerce for personalized recommendations. The market implications include improved efficiency, accuracy, and adaptability in machine learning applications.
Questions about Novel Classifier Training Technology: 1. How does this technology improve the adaptability of classifiers to new data samples? 2. What are the potential commercial uses of this innovation in different industries?
Frequently Updated Research: Stay updated on advancements in classifier training techniques, novel class recognition, and memory-based learning approaches to enhance pattern recognition systems.
Original Abstract Submitted
the present disclosure relates to training a classifier. the classifier includes a controller and an explicit memory. the training may include iteratively receiving one or more second training datasets, each comprising second data samples of a set of one or more associated novel classes, adding to the explicit memory one or more second output vectors indicative of the set of one or more associated novel classes, in response to providing the one or more second training datasets to the classifier, retraining the classifier using the one or more second training datasets and the first training dataset by minimizing a distance between the one or more second output vectors and the one or more prototype vectors, determining a set of updated prototype vectors indicative of first training dataset and the one or more second training datasets, and updating the explicit memory with the set of updated prototype vectors.
