Concept-Oriented Deep Learning
This work addresses foundational issues in machine learning for improving AI systems, but it appears incremental as it builds on existing deep learning frameworks.
The paper tackles the limitations of deep learning in interpretability, transferability, contextual adaptation, and data requirements by proposing concept-oriented deep learning (CODL), which integrates concept representations and understanding, though no concrete results or numbers are provided.
Concepts are the foundation of human deep learning, understanding, and knowledge integration and transfer. We propose concept-oriented deep learning (CODL) which extends (machine) deep learning with concept representations and conceptual understanding capability. CODL addresses some of the major limitations of deep learning: interpretability, transferability, contextual adaptation, and requirement for lots of labeled training data. We discuss the major aspects of CODL including concept graph, concept representations, concept exemplars, and concept representation learning systems supporting incremental and continual learning.