اشترك بالحزمة الذهبية واحصل على وصول غير محدود شمرا أكاديميا
تسجيل مستخدم جديدGradient Descent on Neural Networks Typically Occurs at the Edge of Stability
175
0
0.0
(
0
)
اسأل ChatGPT حول البحث
ﻻ يوجد ملخص باللغة العربية
We empirically demonstrate that full-batch gradient descent on neural network training objectives typically operates in a regime we call the Edge of Stability. In this regime, the maximum eigenvalue of the training loss Hessian hovers just above the numerical value $2 / text{(step size)}$, and the training loss behaves non-monotonically over short timescales, yet consistently decreases over long timescales. Since this behavior is inconsistent with several widespread presumptions in the field of optimization, our findings raise questions as to whether these presumptions are relevant to neural network training. We hope that our findings will inspire future efforts aimed at rigorously understanding optimization at the Edge of Stability. Code is available at https://github.com/locuslab/edge-of-stability.
قيم البحث
اقرأ أيضاً
In this work, we propose to employ information-geometric tools to optimize a graph neural network architecture such as the graph convolutional networks. More specifically, we develop optimization algorithms for the graph-based semi-supervised learnin
g by employing the natural gradient information in the optimization process. This allows us to efficiently exploit the geometry of the underlying statistical model or parameter space for optimization and inference. To the best of our knowledge, this is the first work that has utilized the natural gradient for the optimization of graph neural networks that can be extended to other semi-supervised problems. Efficient computations algorithms are developed and extensive numerical studies are conducted to demonstrate the superior performance of our algorithms over existing algorithms such as ADAM and SGD.
Deep neural networks with batch normalization (BN-DNNs) are invariant to weight rescaling due to their normalization operations. However, using weight decay (WD) benefits these weight-scale-invariant networks, which is often attributed to an increase
of the effective learning rate when the weight norms are decreased. In this paper, we demonstrate the insufficiency of the previous explanation and investigate the implicit biases of stochastic gradient descent (SGD) on BN-DNNs to provide a theoretical explanation for the efficacy of weight decay. We identity two implicit biases of SGD on BN-DNNs: 1) the weight norms in SGD training remain constant in the continuous-time domain and keep increasing in the discrete-time domain; 2) SGD optimizes weight vectors in fully-connected networks or convolution kernels in convolution neural networks by updating components lying in the input feature span, while leaving those components orthogonal to the input feature span unchanged. Thus, SGD without WD accumulates weight noise orthogonal to the input feature span, and cannot eliminate such noise. Our empirical studies corroborate the hypothesis that weight decay suppresses weight noise that is left untouched by SGD. Furthermore, we propose to use weight rescaling (WRS) instead of weight decay to achieve the same regularization effect, while avoiding performance degradation of WD on some momentum-based optimizers. Our empirical results on image recognition show that regardless of optimization methods and network architectures, training BN-DNNs using WRS achieves similar or better performance compared with using WD. We also show that training with WRS generalizes better compared to WD, on other computer vision tasks.
Non-convex optimization problems are challenging to solve; the success and computational expense of a gradient descent algorithm or variant depend heavily on the initialization strategy. Often, either random initialization is used or initialization r
ules are carefully designed by exploiting the nature of the problem class. As a simple alternative to hand-crafted initialization rules, we propose an approach for learning good initialization rules from previous solutions. We provide theoretical guarantees that establish conditions that are sufficient in all cases and also necessary in some under which our approach performs better than random initialization. We apply our methodology to various non-convex problems such as generating adversarial examples, generating post hoc explanations for black-box machine learning models, and allocating communication spectrum, and show consistent gains over other initialization techniques.
Recently there are a considerable amount of work devoted to the study of the algorithmic stability and generalization for stochastic gradient descent (SGD). However, the existing stability analysis requires to impose restrictive assumptions on the bo
undedness of gradients, strong smoothness and convexity of loss functions. In this paper, we provide a fine-grained analysis of stability and generalization for SGD by substantially relaxing these assumptions. Firstly, we establish stability and generalization for SGD by removing the existing bounded gradient assumptions. The key idea is the introduction of a new stability measure called on-average model stability, for which we develop novel bounds controlled by the risks of SGD iterates. This yields generalization bounds depending on the behavior of the best model, and leads to the first-ever-known fast bounds in the low-noise setting using stability approach. Secondly, the smoothness assumption is relaxed by considering loss functions with Holder continuous (sub)gradients for which we show that optimal bounds are still achieved by balancing computation and stability. To our best knowledge, this gives the first-ever-known stability and generalization bounds for SGD with even non-differentiable loss functions. Finally, we study learning problems with (strongly) convex objectives but non-convex loss functions.
We study the complexity of training neural network models with one hidden nonlinear activation layer and an output weighted sum layer. We analyze Gradient Descent applied to learning a bounded target function on $n$ real-valued inputs. We give an agn
ostic learning guarantee for GD: starting from a randomly initialized network, it converges in mean squared loss to the minimum error (in $2$-norm) of the best approximation of the target function using a polynomial of degree at most $k$. Moreover, for any $k$, the size of the network and number of iterations needed are both bounded by $n^{O(k)}log(1/epsilon)$. In particular, this applies to training networks of unbiased sigmoids and ReLUs. We also rigorously explain the empirical finding that gradient descent discovers lower frequency Fourier components before higher frequency components. We complement this result with nearly matching lower bounds in the Statistical Query model. GD fits well in the SQ framework since each training step is determined by an expectation over the input distribution. We show that any SQ algorithm that achieves significant improvement over a constant function with queries of tolerance some inverse polynomial in the input dimensionality $n$ must use $n^{Omega(k)}$ queries even when the target functions are restricted to a set of $n^{O(k)}$ degree-$k$ polynomials, and the input distribution is uniform over the unit sphere; for this class the information-theoretic lower bound is only $Theta(k log n)$. Our approach for both parts is based on spherical harmonics. We view gradient descent as an operator on the space of functions, and study its dynamics. An essential tool is the Funk-Hecke theorem, which explains the eigenfunctions of this operator in the case of the mean squared loss.
الأسئلة المقترحة
سجل دخول لتتمكن من نشر تعليقات
التعليقات
جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
