AI Insight
This theoretical paper presents a unified mathematical framework for understanding learning curves across both biological and artificial systems. The research demonstrates that the shape of learning curves—whether exponential or power-law—is primarily determined by the distribution of situations encountered during learning, with uniform distributions producing exponential curves and power-law distributions yielding power-law curves with predictable exponents. The authors show that mechanisms previously identified in industrial processes like debugging and assembly have been independently rediscovered in deep learning theory as "frequency" and "spectral" scaling laws.
Why it matters
This work provides a fundamental mathematical bridge connecting classical learning theory with modern machine learning, potentially enabling better prediction of training requirements and optimization of learning systems. The finding that different internal mechanisms can produce indistinguishable learning curves has important methodological implications for interpreting experimental results in both cognitive science and artificial intelligence.
Understand the Science
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
Abstract: A general mathematical model of experience formation is considered, in which the learning level is defined as the probability that the next value of the uncertainty factor has already been mastered and not forgotten. Within this model the shape of the learning curve is derived, and it turns out to be determined primarily by the distribution of situation occurrence: a uniform distribution yields an exponential curve, whereas a power-law distribution yields a power-law curve whose exponent equals the ratio of the distribution exponent to that same exponent increased by one. Both mechanisms generating power-law curves, established earlier in the analysis of debugging, assembly and repair activities, are shown to have been independently rediscovered in deep learning theory as the “frequency” and “spectral” mechanisms of scaling laws for artificial neural networks, while the exponential curves of machine learning (contraction operators, strongly convex optimization, stationary reinforcement) are generated by the same constancy of the contraction coefficient that was revealed half a century earlier in models of radioactive decay, the first-order lag element, and others. A solution is given for the problem of choosing a training distribution that differs from the deployment distribution. A methodological consequence is discussed: three distinct mechanisms generate curves indistinguishable in shape, so an observed learning curve does not allow one to judge the “internal” structure of the learning system.
Source: Learning curves in natural and artificial systems