Researchers have introduced a new concept called "learnable novelty" as a unified measure for intelligence across various fields like statistics, computation, and agent behavior. This concept is quantified using a differentiable reservoir computer, which can estimate the portion of surprise that a learner can convert into knowledge. When applied as a measure, it successfully classifies complexity, ranking the Turing-complete rule~110 highest among elementary cellular automata. As an objective, it guides a neural cellular automaton towards complex dynamics and unsupervised image representation, and enhances exploration for reinforcement learning agents. AI
IMPACT Introduces a unified quantitative framework for intelligence, potentially advancing unsupervised learning and agent exploration.
RANK_REASON The item is an academic paper detailing a new theoretical concept and its application. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- CatalyzeX
- Connected Papers
- DagsHub
- Gotit.pub
- Hugging Face
- Litmaps
- MNIST database
- ScienceCast
- scite Smart Citations
- Turing-complete rule~110
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
