arXiv:2605.26973v1 Announce Type: new 
Abstract: Neural networks are known to develop latent representations that are $aligned$, namely structurally similar across networks trained with different architectures, training protocols, or training datasets. We study this phenomenon in a controlled setting, where we train an ensemble of networks on regression and classification tasks using training sets perturbed by independent realizations of a noise process. We show that the signal-to-noise ratio (SNR) and the training sample size influence the alignment in qualitatively similar ways in networks trained on real-world datasets and in an extremely simple $linear$ network with a single hidden layer, for which the alignment can be estimated analytically. Across linear and nonlinear networks, regression and classification tasks, and both synthetic and real-world data, we consistently observe that alignment varies monotonically with SNR but non-monotonically with training sample size. In particular, the alignment is minimized near the interpolation threshold, and a stronger alignment does not necessarily correspond to better generalization error. These findings reveal a non-trivial dependence of alignment on data quality and quantity, decoupled from generalization performance.

دراسة حديثة نُشرت على arXiv تبحث في كيفية تأثير نسبة الإشارة إلى الضوضاء (SNR) وحجم العينة على المحاذاة التمثيلية في الشبكات العصبية. تُظهر الأبحاث أن المحاذاة تتغير باستمرار مع SNR وتظهر علاقة غير أحادية مع حجم العينة عبر هياكل الشبكات المختلفة والمهام.

Un estudio reciente publicado en arXiv investiga cómo la relación señal-ruido (SNR) y el tamaño de la muestra afectan la alineación representacional en redes neuronales. La investigación demuestra que la alineación varía de manera consistente con la SNR y presenta una relación no monótona con el tamaño de la muestra a través de diversas arquitecturas de red y tareas.

Une étude récente publiée sur arXiv examine comment le rapport signal-bruit (SNR) et la taille de l'échantillon influencent l'alignement représentationnel dans les réseaux de neurones. La recherche démontre que l'alignement varie de manière cohérente avec le SNR et présente une relation non monotone avec la taille de l'échantillon à travers diverses architectures de réseau et tâches.

A recent study published on arXiv investigates how the signal-to-noise ratio (SNR) and sample size affect representational alignment in neural networks. The research demonstrates that alignment varies consistently with SNR and exhibits a non-monotonic relationship with sample size across various network architectures and tasks.