Neuron Populations Exhibit Divergent Selectivity with Scale

2026-06-02

AI Summary

This study investigates whether populations of neurons in neural networks exhibit predictable scaling patterns with increasing model size, moving beyond conventional scaling laws that rely solely on macroscopic metrics such as loss. By identifying “Rosetta neurons”—units consistently activated across models—and integrating cross-architecture activation alignment, sublinear power-law fitting, analytical modeling, and data filtering experiments, the work reveals, for the first time, neuron-level scaling laws. Specifically, the number of Rosetta neurons grows sublinearly with model size, while their selectivity and semantic specificity intensify. Concurrently, Rosetta and non-Rosetta neurons undergo polarization: the latter decrease in relative proportion yet become increasingly specialized in domain-specific functions. These findings demonstrate a systematic scaling principle governing the structural organization of neural representations.

Abstract

We investigate whether neuron populations within neural networks evolve predictably with scale, extending scaling laws beyond macroscopic observables such as loss. To probe this question, we study Rosetta Neurons, a previously characterized class of neurons whose activation patterns are similar across independently trained models (Dravid et al., 2023). In separate analyses of language models up to 30B parameters and vision models up to 5B parameters, we observe that the population of Rosetta Neurons follows a sublinear power law in model size, growing in absolute number but occupying a shrinking fraction of the total neuron count. We further observe a Neuron Polarization Effect: Rosetta Neurons become more selective and increasingly monosemantic with scale, separating from a growing non-Rosetta population that remains less selective. An analytical model balancing feature utility against limited neuron capacity explains the sublinear power-law scaling and this polarization effect. Finally, we find that Rosetta Neurons become more domain-specialized with scale and illustrate their selectivity through a targeted data-filtering case study for continued pretraining. Our results point to a scaling law for interpretable, shared neuron-level structure, linking model size to systematic changes in neuron universality, selectivity, and specialization.

Problem

• scaling laws
• neuron selectivity
• Rosetta Neurons
• model scale
• neuron specialization

Innovation

• scaling laws
• Rosetta Neurons
• neuron selectivity
• monosemanticity
• neuron polarization