{"id":"arxiv/2605.14567","url":"https://arxiv.org/abs/2605.14567","tags":["stat.ML","cs.LG","math.PR","math.ST"],"score":0,"title":"Scaling Laws from Sequential Feature Recovery: A Solvable Hierarchical Model","domain":"stat.ML","source":"arxiv","authors":["Arie Wortsman-Zurich","Hugo Tabanelli","Yatin Dandi","Florent Krzakala","Bruno Loureiro"],"reviews":0,"subject":"stat.ML","summary":"We propose a simple mechanism by which scaling laws emerge from feature learning in multi-layer networks. We study a high-dimensional hierarchical target that is a globally high-degree function, but that can be represented by a combination of latent compositional features whose weights decrease as a power law. We show that a layer-wise spectral algorithm adapted to this compositional structure achieves improved scaling relative to shallow, non-adaptive methods, and recovers the latent directions sequentially: strong features become detectable at small sample sizes, while weaker features…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-05-14","reactions":{}}