Watch this video - A Research Programme for Continual and Neurosymbolic Epistemic AI

This research proposes a novel framework called Epistemic Artificial Intelligence to address the inherent brittleness of modern deep learning models. The authors argue that current AI struggles with epistemic uncertainty, which arises from a lack of knowledge about the world and leads to unreliable predictions when models encounter unfamiliar data. To solve this, they advocate for a continual learning paradigm that uses second-order mathematical theories, such as belief functions and random sets, to provide safety guarantees throughout a machine's lifespan. The program also introduces epistemic neurosymbolic AI, a method that integrates human-like symbolic reasoning with neural networks to reduce uncertainty through logic and background knowledge. By combining these approaches, the researchers aim to create more robust and transparent systems capable of handling complex tasks like autonomous driving and scene understanding. Ultimately, this work seeks to establish a unified mathematical foundation for AI that explicitly acknowledges and manages what a model does not know

#EpistemicArtificialIntelligence, #EpistemicUncertainty, #ContinualLearning, #NeurosymbolicAI, #EpistemicNeurosymbolicAI, #BeliefFunctions, #RandomSets, #SecondOrderMathematics, #RobustAI, #TransparentAI, #SafeAI, #TrustworthyAI, #MachineLearning, #DeepLearning, #AIResearch, #ArtificialIntelligence, #AutonomousDriving, #SceneUnderstanding, #Robustness, #UncertaintyQuantification, #MathematicalFoundation, #AIExplainers, #SafetyGuarantees, #HumanLikeReasoning, #SymbolicReasoning, #NeuralNetworks, #LogicAndAI, #KnowledgeBasedAI, #FutureOfAI, #NextGenAI

Видео Watch this video - A Research Programme for Continual and Neurosymbolic Epistemic AI канала CONVERGENZE POSSIBILI di Stefano Donno