Metacognitive AI: Recovering Constraints by Finding ML Errors

Explore a cutting-edge approach to improving machine learning models through metacognitive AI in this 18-minute video presentation. Delve into the process of learning rules to detect and correct errors in ML models, with a focus on the latest research presented in a CIKM'24 paper. Discover how identifying class combinations that lead to errors in multi-class settings can function as constraints, and learn how these constraints are applied using neurosymbolic methods like LTN to enhance model performance. Gain insights into the intersection of symbolic methods and deep learning, and understand the potential implications for advancing artificial general intelligence (AGI). Access the full research paper and expand your knowledge of this exciting area in artificial intelligence and machine learning.