algorithms that improve through experience
in cyber, the cybergraph itself is a learning system. neurons create cyberlinks, the tru computes focus, and the resulting knowledge emerges from collective learning
the learning loop:
- neurons observe the current focus distribution
- they submit new cyberlinks to increase relevance of valuable particles
- the tri-kernel recomputes ranks, redistributing attention
- karma rewards neurons whose contributions align with collective judgment
this is a decentralized gradient descent on the relevance landscape. each cyberlink is a training sample. the cybergraph is the model. focus is the loss signal
machine learning methods also appear inside cyber's stack: neural language models for inference, embedding models for semantic similarity, reinforcement learning for agent behavior
the distinction between classical ML and cyber's collective learning: in classical ML one entity owns the model. in cyber, the model is the shared cybergraph and every neuron is a trainer
see learning, cybergraph, tru, focus, neural language, karma