More Like this

1. Learning in the Presence of Low-dimensional Structure: A Spiked Random Matrix Perspective

   Ba, Jimmy; Erdogdu, Murat A; Suzuki, Taiji; Wang, Zhichao; Wu, Denny (December 2023, Advances in Neural Information Processing Systems 36 (NeurIPS 2023))

   Oh, A; Naumann, T; Globerson, A; Saenko, K; Hardt, M; Levine, S (Ed.)

   We consider the problem of learning a single-index target function f∗ : Rd → R under the spiked covariance data: f∗(x) = σ∗   √ 1 1+θ ⟨x,μ⟩   , x ∼ N(0, Id + θμμ⊤), θ ≍ dβ for β ∈ [0, 1), where the link function σ∗ : R → R is a degree-p polynomial with information exponent k (defined as the lowest degree in the Hermite expansion of σ∗), and it depends on the projection of input x onto the spike (signal) direction μ ∈ Rd. In the proportional asymptotic limit where the number of training examples n and the dimensionality d jointly diverge: n, d → ∞, n/d → ψ ∈ (0,∞), we ask the following question: how large should the spike magnitude θ be, in order for (i) kernel methods, (ii) neural networks optimized by gradient descent, to learn f∗? We show that for kernel ridge regression, β ≥ 1 − 1 p is both sufficient and necessary. Whereas for two-layer neural networks trained with gradient descent, β > 1 − 1 k suffices. Our results demonstrate that both kernel methods and neural networks benefit from low-dimensional structures in the data. Further, since k ≤ p by definition, neural networks can adapt to such structures more effectively. 

   more » « less
2. Statistical and Computational Complexities of BFGS Quasi-Newton Method for Generalized Linear Models

   Jin, Qiujiang; Ren, Tongzheng; Ho, Nhat; Mokhtari, Aryan (May 2024, Transactions on machine learning research)

   The gradient descent (GD) method has been used widely to solve parameter estimation in generalized linear models (GLMs), a generalization of linear models when the link function can be non-linear. In GLMs with a polynomial link function, it has been shown that in the high signal-to-noise ratio (SNR) regime, due to the problem's strong convexity and smoothness, GD converges linearly and reaches the final desired accuracy in a logarithmic number of iterations. In contrast, in the low SNR setting, where the problem becomes locally convex, GD converges at a slower rate and requires a polynomial number of iterations to reach the desired accuracy. Even though Newton's method can be used to resolve the flat curvature of the loss functions in the low SNR case, its computational cost is prohibitive in high-dimensional settings as it is $$\mathcal{O}(d^3)$$, where $$d$$ the is the problem dimension. To address the shortcomings of GD and Newton's method, we propose the use of the BFGS quasi-Newton method to solve parameter estimation of the GLMs, which has a per iteration cost of $$\mathcal{O}(d^2)$$. When the SNR is low, for GLMs with a polynomial link function of degree $$p$$, we demonstrate that the iterates of BFGS converge linearly to the optimal solution of the population least-square loss function, and the contraction coefficient of the BFGS algorithm is comparable to that of Newton's method. Moreover, the contraction factor of the linear rate is independent of problem parameters and only depends on the degree of the link function $$p$$. Also, for the empirical loss with $$n$$ samples, we prove that in the low SNR setting of GLMs with a polynomial link function of degree $$p$$, the iterates of BFGS reach a final statistical radius of $$\mathcal{O}((d/n)^{\frac{1}{2p+2}})$$ after at most $$\log(n/d)$$ iterations. This complexity is significantly less than the number required for GD, which scales polynomially with $(n/d)$. 

   more » « less
3. Temporal Difference Learning as Gradient Splitting

   Liu, Rui; Olshevsky, Alex (November 2021, International Conference on Machine Learning (ICML))

   Temporal difference learning with linear function approximation is a popular method to obtain a low-dimensional approximation of the value function of a policy in a Markov Decision Process. We give a new interpretation of this method in terms of a splitting of the gradient of an appropriately chosen function. As a consequence of this interpretation, convergence proofs for gradient descent can be applied almost verbatim to temporal difference learning. Beyond giving a new, fuller explanation of why temporal difference works, our interpretation also yields improved convergence times. We consider the setting with 1/T^{1/2} step-size, where previous comparable finite-time convergence time bounds for temporal difference learning had the multiplicative factor 1/(1-\gamma) in front of the bound, with γ being the discount factor. We show that a minor variation on TD learning which estimates the mean of the value function separately has a convergence time where 1/(1-\gamma) only multiplies an asymptotically negligible term. 

   more » « less
4. Symmetric Single-Index Learning

   Zweig, Aaron; Bruna, Joan (May 2024, International Conference on Learning Representations (ICLR))

   Few neural architectures lend themselves to provable learning with gradient based methods. One popular model is the single-index model, in which labels are produced by composing an unknown linear projection with a possibly unknown scalar link function. Learning this model with SGD is relatively well-understood, whereby the so-called information exponent of the link function governs a polynomial sample complexity rate. However, extending this analysis to deeper or more complicated architectures remains challenging. In this work, we consider single index learning in the setting of symmetric neural net- works. Under analytic assumptions on the activation and maximum degree assumptions on the link function, we prove that gradient flow recovers the hidden planted direction, represented as a finitely supported vector in the feature space of power sum polynomials. We characterize a notion of information exponent adapted to our setting that controls the efficiency of learning. 

   more » « less
5. Self-Stabilization: The Implicit Bias of Gradient Descent at the Edge of Stability.

   Damian, Alex; Nichani, Eshaan; Lee, Jason D. (May 2023, ICLR 2023)

   Traditional analyses of gradient descent show that when the largest eigenvalue of the Hessian, also known as the sharpness S(θ), is bounded by 2/η, training is "stable" and the training loss decreases monotonically. Recent works, however, have observed that this assumption does not hold when training modern neural networks with full batch or large batch gradient descent. Most recently, Cohen et al. (2021) observed two important phenomena. The first, dubbed progressive sharpening, is that the sharpness steadily increases throughout training until it reaches the instability cutoff 2/η. The second, dubbed edge of stability, is that the sharpness hovers at 2/η for the remainder of training while the loss continues decreasing, albeit non-monotonically. We demonstrate that, far from being chaotic, the dynamics of gradient descent at the edge of stability can be captured by a cubic Taylor expansion: as the iterates diverge in direction of the top eigenvector of the Hessian due to instability, the cubic term in the local Taylor expansion of the loss function causes the curvature to decrease until stability is restored. This property, which we call self-stabilization, is a general property of gradient descent and explains its behavior at the edge of stability. A key consequence of self-stabilization is that gradient descent at the edge of stability implicitly follows projected gradient descent (PGD) under the constraint S(θ)≤2/η. Our analysis provides precise predictions for the loss, sharpness, and deviation from the PGD trajectory throughout training, which we verify both empirically in a number of standard settings and theoretically under mild conditions. Our analysis uncovers the mechanism for gradient descent's implicit bias towards stability. 

   more » « less