A5082444106- Intelligent Tutoring Systems and Adaptive Learning
- Machine Learning and ELM
- Domain Adaptation and Few-Shot Learning
- Education and Learning Interventions
- Machine Learning and Algorithms
- Innovative Teaching and Learning Methods
- Data Stream Mining Techniques
- Machine Learning and Data Classification
- Online Learning and Analytics
Commissariat à l'Énergie Atomique et aux Énergies Alternatives
2023-2024
CEA LIST
2023-2024
Université Paris-Saclay
2023
CEA Paris-Saclay
2023
Class-Incremental Learning (CIL) aims to build classification models from data streams. At each step of the CIL process, new classes must be integrated into model. Due catastrophic forgetting, is particularly challenging when examples past cannot stored, case on which we focus here. To date, most approaches are based exclusively target dataset process. However, use pre-trained in a self-supervised way large amounts has recently gained momentum. The initial model process may only first batch...
Recent class-incremental learning methods combine deep neural architectures and algorithms to handle streaming data under memory computational constraints. The performance of existing varies depending on the characteristics incremental process. To date, there is no other approach than test all pairs training available at start process select a suited algorithm-architecture combination. tackle this problem, in article, we introduce AdvisIL, method which takes as input main (memory budget for...
Class-incremental learning deals with sequential data streams composed of batches classes. Various algorithms have been proposed to address the challenging case where samples from past classes cannot be stored. However, selecting an appropriate algorithm for a user-defined setting is open problem, as relative performance these depends on incremental settings. To solve this we introduce recommendation method that simulates future stream. Given initial set classes, it leverages generative...
Class-Incremental Learning (CIL) aims to build classification models from data streams. At each step of the CIL process, new classes must be integrated into model. Due catastrophic forgetting, is particularly challenging when examples past cannot stored, case on which we focus here. To date, most approaches are based exclusively target dataset process. However, use pre-trained in a self-supervised way large amounts has recently gained momentum. The initial model process may only first batch...