# Neural Scaling Laws **Domain:** Machine Learning / Deep Learning / Empirical Modeling **Doc Type:** Technical Concept Node **Maturity:** Developing ## Definition **Neural scaling laws** describe regular empirical relationships between model performance and quantities such as parameter count, training data and computational budget across ranges of scale. ## Historical Context Researchers in the 1970s understood that more computation, memory, units and data could improve machine intelligence. What they lacked was the later mapped empirical regime showing comparatively regular performance changes across orders of magnitude under relatively stable architectures and training procedures. ## Evidence Boundary Scaling curves describe observed relationships. They do not by themselves explain why a capability emerges, guarantee that a trend will continue, or establish whether an apparent threshold is intrinsic or measurement-dependent. ## See Also [[wiki/Deep Learning|Deep Learning]], [[wiki/Artificial Neural Networks|Artificial Neural Networks]], [[wiki/AI Benchmarking|AI Benchmarking]]