KC-Agent: A Dual-Process Cognitive Architecture for Efficient ML Model Improvement
A recent study presents KC-Agent, a cognitive architecture that employs a dual-process approach to enhance machine learning models in operational settings. This system merges rapid pattern recognition (System 1) with methodical incremental adjustments (System 2), utilizing structured memory to capitalize on effective solutions and minimize expensive recalculations. It features principles of atomic changes and rollback functions for dependable updates. Tested across five datasets, including real temporal degradation from NASA turbofan data and synthetic drift scenarios, KC-Agent reaches an impressive accuracy of 76.8%. The research can be accessed on arXiv (2608.02351).
Key facts
- KC-Agent is a dual-process cognitive architecture for automated ML model improvement.
- It combines System 1 (fast pattern recognition) and System 2 (deliberate incremental updates).
- Structured memory enables System 1 to reuse solutions discovered by System 2.
- Atomic change principles and rollback capabilities ensure reliable updates.
- Evaluated on five datasets, including NASA turbofan data with real temporal degradation.
- Achieves state-of-the-art performance with 76.8% accuracy.
- Paper announced on arXiv with ID 2608.02351.
Entities
Institutions
- arXiv