High-dimensional Asymptotics of Feature Learning: How One Gradient Step Improves the Representation

• URL: http://arxiv.org/abs/2205.01445v1
• Date: Tue, 3 May 2022 12:09:59 GMT
• Title: High-dimensional Asymptotics of Feature Learning: How One Gradient Step Improves the Representation
• Authors: Jimmy Ba, Murat A. Erdogdu, Taiji Suzuki, Zhichao Wang, Denny Wu, Greg Yang
• Abstract summary: 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.
• Score: 89.21686761957383
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: We study the first gradient descent step on the first-layer parameters $\boldsymbol{W}$ in a two-layer neural network: $f(\boldsymbol{x}) = \frac{1}{\sqrt{N}}\boldsymbol{a}^\top\sigma(\boldsymbol{W}^\top\boldsymbol{x})$, where $\boldsymbol{W}\in\mathbb{R}^{d\times N}, \boldsymbol{a}\in\mathbb{R}^{N}$ are randomly initialized, and the training objective is the empirical MSE loss: $\frac{1}{n}\sum_{i=1}^n (f(\boldsymbol{x}_i)-y_i)^2$. In the proportional asymptotic limit where $n,d,N\to\infty$ at the same rate, and an idealized student-teacher setting, we show that the first gradient update contains a rank-1 "spike", which results in an alignment between the first-layer weights and the linear component of the teacher model $f^*$. To characterize the impact of this alignment, we compute the prediction risk of ridge regression on the conjugate kernel after one gradient step on $\boldsymbol{W}$ with learning rate $\eta$, when $f^*$ is a single-index model. We consider two scalings of the first step learning rate $\eta$. For small $\eta$, we establish a Gaussian equivalence property for the trained feature map, and prove that the learned kernel improves upon the initial random features model, but cannot defeat the best linear model on the input. Whereas for sufficiently large $\eta$, we prove that for certain $f^*$, the same ridge estimator on trained features can go beyond this "linear regime" and outperform a wide range of random features and rotationally invariant kernels. Our results demonstrate that even one gradient step can lead to a considerable advantage over random features, and highlight the role of learning rate scaling in the initial phase of training.

Related papers

• A Unified Framework for Uniform Signal Recovery in Nonlinear Generative Compressed Sensing [68.80803866919123]
  Under nonlinear measurements, most prior results are non-uniform, i.e., they hold with high probability for a fixed $mathbfx*$ rather than for all $mathbfx*$ simultaneously. Our framework accommodates GCS with 1-bit/uniformly quantized observations and single index models as canonical examples. We also develop a concentration inequality that produces tighter bounds for product processes whose index sets have low metric entropy.
  arXiv  Detail & Related papers  (2023-09-25T17:54:19Z)
• Solving Quadratic Systems with Full-Rank Matrices Using Sparse or Generative Priors [44.94521933974231]
  This paper addresses the problem of recovering a signal from a quadratic system $y_i=boldsymbolxtopboldsymbolA_iboldsymbolx. We introduce the thresholded Wirtinger flow (TWF) algorithm that does not require the sparsity level $k$. We show that our approach for the sparse case notably outperforms the existing provable algorithm power factorization.
  arXiv  Detail & Related papers  (2023-09-16T16:00:07Z)
• Depth Dependence of $\mu$P Learning Rates in ReLU MLPs [72.14317069090407]
  We study the dependence on $n$ and $L$ of the maximal update ($mu$P) learning rate. We find that it has a non-trivial dependence of $L$, scaling like $L-3/2.$
  arXiv  Detail & Related papers  (2023-05-13T01:10:49Z)
• Generalization and Stability of Interpolating Neural Networks with Minimal Width [37.908159361149835]
  We investigate the generalization and optimization of shallow neural-networks trained by gradient in the interpolating regime. We prove the training loss number minimizations $m=Omega(log4 (n))$ neurons and neurons $Tapprox n$. With $m=Omega(log4 (n))$ neurons and $Tapprox n$, we bound the test loss training by $tildeO (1/)$.
  arXiv  Detail & Related papers  (2023-02-18T05:06:15Z)
• 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)
• Neural Networks can Learn Representations with Gradient Descent [68.95262816363288]
  In specific regimes, neural networks trained by gradient descent behave like kernel methods. In practice, it is known that neural networks strongly outperform their associated kernels.
  arXiv  Detail & Related papers  (2022-06-30T09:24:02Z)
• Nearly Minimax Optimal Reinforcement Learning with Linear Function Approximation [25.60689712525918]
  We study reinforcement learning with linear function approximation where the transition probability and reward functions are linear. We propose a novel-efficient algorithm, LSVI-UCB$+$, which achieves an $widetildeO(HdsqrtT)$ regret bound where $H$ is the episode length, $d$ is the feature dimension, and $T$ is the number of steps.
  arXiv  Detail & Related papers  (2022-06-23T06:04:21Z)
• Universality of empirical risk minimization [12.764655736673749]
  Consider supervised learning from i.i.d. samples where $boldsymbol x_i inmathbbRp$ are feature vectors and $y in mathbbR$ are labels. We study empirical risk universality over a class of functions that are parameterized by $mathsfk.
  arXiv  Detail & Related papers  (2022-02-17T18:53:45Z)
• Beyond Lazy Training for Over-parameterized Tensor Decomposition [69.4699995828506]
  We show that gradient descent on over-parametrized objective could go beyond the lazy training regime and utilize certain low-rank structure in the data. Our results show that gradient descent on over-parametrized objective could go beyond the lazy training regime and utilize certain low-rank structure in the data.
  arXiv  Detail & Related papers  (2020-10-22T00:32:12Z)
• 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.