Phase Transitions for Feature Learning in Neural Networks

• URL: http://arxiv.org/abs/2602.01434v1
• Date: Sun, 01 Feb 2026 20:47:36 GMT
• Title: Phase Transitions for Feature Learning in Neural Networks
• Authors: Andrea Montanari, Zihao Wang,
• Abstract summary: We study the descent dynamics of two-layer neural networks under the proportional neuronss $n,dtoin$, $n/dto$.<n>Our characterization of $_textNN$ opens the way to study the dependence of learning dynamics on the network architecture and training algorithm.
• Score: 27.411134657066267
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: According to a popular viewpoint, neural networks learn from data by first identifying low-dimensional representations, and subsequently fitting the best model in this space. Recent works provide a formalization of this phenomenon when learning multi-index models. In this setting, we are given $n$ i.i.d. pairs $({\boldsymbol x}_i,y_i)$, where the covariate vectors ${\boldsymbol x}_i\in\mathbb{R}^d$ are isotropic, and responses $y_i$ only depend on ${\boldsymbol x}_i$ through a $k$-dimensional projection ${\boldsymbol Θ}_*^{\sf T}{\boldsymbol x}_i$. Feature learning amounts to learning the latent space spanned by ${\boldsymbol Θ}_*$. In this context, we study the gradient descent dynamics of two-layer neural networks under the proportional asymptotics $n,d\to\infty$, $n/d\toδ$, while the dimension of the latent space $k$ and the number of hidden neurons $m$ are kept fixed. Earlier work establishes that feature learning via polynomial-time algorithms is possible if $δ> δ_{\text{alg}}$, for $δ_{\text{alg}}$ a threshold depending on the data distribution, and is impossible (within a certain class of algorithms) below $δ_{\text{alg}}$. Here we derive an analogous threshold $δ_{\text{NN}}$ for two-layer networks. Our characterization of $δ_{\text{NN}}$ opens the way to study the dependence of learning dynamics on the network architecture and training algorithm. The threshold $δ_{\text{NN}}$ is determined by the following scenario. Training first visits points for which the gradient of the empirical risk is large and learns the directions spanned by these gradients. Then the gradient becomes smaller and the dynamics becomes dominated by negative directions of the Hessian. The threshold $δ_{\text{NN}}$ corresponds to a phase transition in the spectrum of the Hessian in this second phase.

Related papers

• Neural Networks Learn Generic Multi-Index Models Near Information-Theoretic Limit [66.20349460098275]
  We study the gradient descent learning of a general Gaussian Multi-index model $f(boldsymbolx)=g(boldsymbolUboldsymbolx)$ with hidden subspace $boldsymbolUin mathbbRrtimes d$.<n>We prove that under generic non-degenerate assumptions on the link function, a standard two-layer neural network trained via layer-wise gradient descent can agnostically learn the target with $o_d(1)$ test error.
  arXiv  Detail & Related papers  (2025-11-19T04:46:47Z)
• Emergence and scaling laws in SGD learning of shallow neural networks [64.48316762675141]
  We study the complexity of online gradient descent (SGD) for learning a two-layer neural network with $P$ neurons on isotropic Gaussian data.<n>We provide a precise analysis of SGD dynamics for the training of a student two-layer network to minimize the mean squared error (MSE) objective.
  arXiv  Detail & Related papers  (2025-04-28T16:58:55Z)
• Neural network learns low-dimensional polynomials with SGD near the information-theoretic limit [75.4661041626338]
  We study the problem of gradient descent learning of a single-index target function $f_*(boldsymbolx) = textstylesigma_*left(langleboldsymbolx,boldsymbolthetarangleright)$<n>We prove that a two-layer neural network optimized by an SGD-based algorithm learns $f_*$ with a complexity that is not governed by information exponents.
  arXiv  Detail & Related papers  (2024-06-03T17:56:58Z)
• Learning Hierarchical Polynomials with Three-Layer Neural Networks [56.71223169861528]
  We study the problem of learning hierarchical functions over the standard Gaussian distribution with three-layer neural networks. For a large subclass of degree $k$s $p$, a three-layer neural network trained via layerwise gradientp descent on the square loss learns the target $h$ up to vanishing test error. This work demonstrates the ability of three-layer neural networks to learn complex features and as a result, learn a broad class of hierarchical functions.
  arXiv  Detail & Related papers  (2023-11-23T02:19:32Z)
• Solving Quadratic Systems with Full-Rank Matrices Using Sparse or Generative Priors [33.0212223058894]
  The problem of recovering a signal from a quadratic system $y_i=boldsymbol xtopboldsymbol A_iboldsymbol x, i=1,ldots,m$ with full-rank matrices $boldsymbol A_i$ frequently arises in applications such as unassigned distance geometry and sub-wavelength imaging. This paper addresses the high-dimensional case where $mll n$ incorporating by prior knowledge of $boldsymbol x$.
  arXiv  Detail & Related papers  (2023-09-16T16:00:07Z)
• Neural Networks Efficiently Learn Low-Dimensional Representations with SGD [22.703825902761405]
  We show that SGD-trained ReLU NNs can learn a single-index target of the form $y=f(langleboldsymbolu,boldsymbolxrangle) + epsilon$ by recovering the principal direction. We also provide compress guarantees for NNs using the approximate low-rank structure produced by SGD.
  arXiv  Detail & Related papers  (2022-09-29T15:29:10Z)
• High-dimensional Asymptotics of Feature Learning: How One Gradient Step Improves the Representation [89.21686761957383]
  We study the first gradient descent step on the first-layer parameters $boldsymbolW$ in a two-layer network. Our results demonstrate that even one step can lead to a considerable advantage over random features.
  arXiv  Detail & Related papers  (2022-05-03T12:09:59Z)
• Learning Over-Parametrized Two-Layer ReLU Neural Networks beyond NTK [58.5766737343951]
  We consider the dynamic of descent for learning a two-layer neural network. We show that an over-parametrized two-layer neural network can provably learn with gradient loss at most ground with Tangent samples.
  arXiv  Detail & Related papers  (2020-07-09T07:09:28Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.