Theoretical analysis explains why models generalize after overfitting in modular addition.
problem Understanding why models generalize after overfitting in modular addition.
method Theoretical analysis and gradient descent behavior of two-layer quadratic networks and Transformers.
result Two-layer quadratic networks and simple Transformers generalize well after initially overfitting, indicating grokking.
New findings on kernel regression in the quadratic regime, improving understanding of machine learning models.
problem Understanding kernel ridge regression in the quadratic asymptotic regime.
method Extended study of kernel regression to the quadratic regime, establishing approximation bounds and spectral distributions.
result Broad class of inner-product kernels exhibit behavior similar to a quadratic kernel, with precise asymptotic training and test errors characterized.
A recent line of work studies overparametrized neural networks in the "kernel regime," i.e. when the network behaves during training as a kernelized linear predictor, and thus training with gradient descent has the effect of finding the minimum RKHS norm solution. This stands in contrast to other studies which demonstr…
A recent line of work studies overparametrized neural networks in the "kernel regime," i.e. when the network behaves during training as a kernelized linear predictor, and thus training with gradient descent has the effect of finding the minimum RKHS norm solution. This stands in contrast to other studies which demonstr…
Averaged SGD achieves optimal convergence rate for neural networks in the NTK regime.
problem Convergence analysis of averaged stochastic gradient descent for neural networks.
method Analyzed convergence of averaged stochastic gradient descent for overparameterized two-layer neural networks.
result Achieved minimax optimal convergence rate with global convergence guarantee.
Recent work by Jacot et al. (2018) has shown that training a neural network using gradient descent in parameter space is related to kernel gradient descent in function space with respect to the Neural Tangent Kernel (NTK). Lee et al. (2019) built on this result by establishing that the output of a neural network traine…
Gradient descent learns useful features even in the NTK regime.
problem The ability of neural networks to learn useful features.
method Local convergence analysis of gradient descent with regularization.
result Gradient descent can capture ground-truth directions for feature learning even after the loss threshold is reached.
Model shows feature learning can improve neural scaling laws for hard tasks.
problem Understanding and improving neural network scaling laws for various task difficulties.
method Developed a solvable model of neural scaling laws, identified three scaling regimes, and demonstrated feature learning's impact on scaling exponents.
result Feature learning can improve scaling with training time and compute for hard tasks, nearly doubling the exponent.
This paper studies the optimality of kernel methods in high-dimensional data clustering. Recent works have studied the large sample performance of kernel clustering in the high-dimensional regime, where Euclidean distance becomes less informative. However, it is unknown whether popular methods, such as kernel k-means, …
Knowledge transfer speeds up neural classifier training.
problem Lack of theoretical analysis of knowledge transfer in neural networks.
method Regularization of fit between teacher and student networks using privileged information.
result Wide two-layer networks can interpolate between privileged information and data, improving generalization.
The paper analyzes learning curves for kernel ridge regression with dot-product kernels.
problem Understanding the learning curves for different scaling regimes of data and model.
method Precise formulas for mean test error, bias, and variance in the mo∞ with m/dr constant regime. result A peak in the learning curve at m≈dr/r! for any integer r. Analyzes neural networks using linear models to understand their behavior.
problem Understanding multi-layer neural networks through linear models.
method Recalls and reviews four models: linear regression with concentrated features, kernel ridge regression, random feature model, and neural tangent model.
result Highlights limitations of linear theory and discusses approaches to overcome them.
Study reveals learning curves and benign overfitting in spectral algorithms for large dimensions.
problem Understanding learning curves and benign overfitting in spectral algorithms for large-dimensional data.
method Analysis of learning curves and benign overfitting in spectral algorithms for inner-product kernels on the sphere and general domains.
result Characterization of three distinct regimes: over-regularized, under-regularized, and interpolation regimes, revealing benign overfitting across both under-regularized and interpolation regimes.
The paper studies how neural networks evolve representations, finding a unique fixed point for nonlinear activations.
problem Understanding how neural networks transform input data across layers.
method Theoretical framework for the evolution of the kernel sequence, using mean-field regime and Hermite polynomials.
result For nonlinear activations, the kernel sequence converges globally to a unique fixed point.
Kernel methods and MLPs perform similarly to linear models in high dimensions.
problem Understanding the performance of kernel methods and MLPs in high-dimensional settings.
method Analysis of kernel methods and MLPs in a high-dimensional regime with proportional asymptotics.
result Linear models are optimal in high-dimensional settings when data is generated by kernel models with nonlinear relationships.
Two-layer neural networks learn efficiently using kernel methods in mean-field analysis.
problem Feature learning ability of two-layer neural networks in the mean-field regime.
method Mean-field analysis through kernel methods, focusing on dynamics of the first layer's kernel.
result Two-layer neural networks can learn a union of multiple reproducing kernel Hilbert spaces more efficiently than kernel methods.
Embedded ensembles improve neural network performance efficiently.
problem Improving neural network performance with fewer resources.
method Analyzing the wide network limit of gradient descent dynamics using Neural-Tangent-Kernel.
result Embedded ensembles exhibit two regimes: independent and collective, affecting performance.
New method preserves privacy while improving machine learning accuracy.
problem Privacy-preserving machine learning for daily data.
method Compressive Privacy and multi-kernel method.
result Improved utility classification accuracy with privacy preservation.
A fairly comprehensive analysis is presented for the gradient descent dynamics for training two-layer neural network models in the situation when the parameters in both layers are updated. General initialization schemes as well as general regimes for the network width and training data size are considered. In the over-…
New approach predicts generalization of deep neural networks in proportional-width regime.
problem Predicting generalization of deep neural networks in proportional-width regime.
method Equivalent Wishart Ansatz for hierarchical empirical kernels, renormalized NNGP kernel.
result Renormalized NNGP kernel captures dominant stochastic fluctuations in deep neural networks.
Linear RNNs exhibit a bias towards shorter memory due to initialization variance.
problem Understanding the performance limitations of RNNs, especially linear ones.
method Kernel regime analysis to show equivalence to 1D-convolutional networks and analyze weightings.
result Linear RNNs with random initialization have a bias towards shorter memory periods.
The study analyzes how many neurons are needed for two-layer neural networks trained with gradient descent.
problem Determining the minimum number of neurons required for effective training of shallow neural networks.
method Analyzes two-layer neural networks in the NTK regime, trained with gradient descent. Derives fast rates of convergence and tracks the number of hidden neurons required for generalization.
result Derives fast rates of convergence and improves on existing results for the number of hidden neurons needed for generalization.
This article proposes a performance analysis of kernel least squares support vector machines (LS-SVMs) based on a random matrix approach, in the regime where both the dimension of data p and their number n grow large at the same rate. Under a two-class Gaussian mixture model for the input data, we prove that the LS…
The paper analyzes Kernel Density Estimation in high dimensions with varying data and dimensionality.
problem High-dimensional Kernel Density Estimation with growing data and dimensionality.
method Examines the behavior of Kernel Density Estimators in the regime where both data points and dimensionality grow with a fixed ratio.
result Three distinct statistical regimes are identified for Kernel-based density estimates, each with different statistical properties.
DeRegiME forecasts with regime structure, improving probabilistic predictions across various time series.
problem Probabilistic forecasting discards residual uncertainty, and distribution shifts are hard to capture.
method DeRegiME uses a sparse variational Gaussian process with a nonstationary regime-mixing kernel to separate latent uncertainty regimes.
result DeRegiME improves NLPD by 20.3% on average across benchmarks, with gains on CRPS and MSE.
Study on learning properties of scale-dependent kernels controlling stability and error.
problem Understanding the learning properties of scale-dependent kernels in nonparametric ridge-less least squares.
method Combines probabilistic results with interpolation theory to analyze stability and error.
result Different regimes of learning error depending on sample size and data dimension.
The paper analyzes the stationarity of stochastic Volterra integral equations and introduces fake stationary regimes.
problem Analyzing the stationarity of non-Markovian dynamical systems described by SVIEs.
method Investigates the properties of SVIE solutions, focusing on stationarity over finite and long time horizons, and introduces a deterministic stabilizer to induce a fake stationary regime.
result SVIEs do not exhibit a strong stationary regime unless the kernel is constant or degenerate, but a fake stationary regime can be achieved with a deterministic stabilizer.
Adaptive kernels from neural networks improve model performance.
problem Improving neural network performance through adaptive kernels.
method Deriving adaptive kernels from infinite-width neural networks using feature learning and gradient flow training.
result Adaptive kernels achieve lower test loss compared to traditional kernels.
The paper analyzes high-dimensional kernel regression, showing different risk curves based on data and regularization.
problem Characterizing generalization properties of high-dimensional kernel ridge regression.
method Bias-variance decomposition of the expected excess risk, considering different regularization schemes and data eigen-profiles.
result The risk curve of kernel regression can be double-descent-like, bell-shaped, or monotonic, depending on n, d, and regularization level.
This paper carries out a large dimensional analysis of a variation of kernel ridge regression that we call \emph{centered kernel ridge regression} (CKRR), also known in the literature as kernel ridge regression with offset. This modified technique is obtained by accounting for the bias in the regression problem resulti…
Neural networks generalize well despite overfitting due to high capacity.
problem Understanding why deep neural networks generalize well in overparameterized settings.
method High-dimensional asymptotic analysis of generalization under kernel regression with Neural Tangent Kernel.
result Test error exhibits non-monotonic behavior and can have additional peaks and descents in the overparameterized regime.
Deep ReLU networks approximate as well as shallow ones in kernel regimes.
problem Understanding the limitations of kernel methods for deep ReLU networks.
method Characterizing eigenvalue decays of kernels derived from deep ReLU networks.
result Deep ReLU networks and shallow two-layer networks have equivalent approximation properties in kernel regimes.
This paper analyzes error in SKI for Gaussian Processes, providing conditions for linear time inference.
problem Lack of rigorous theoretical error analysis for SKI.
method Proved error bounds for SKI Gram matrix, examined error effects, provided practical guidelines.
result Identified two dimensionality regimes for SKI's scalability-accuracy trade-offs.
Linearized attention fails to converge to NTK limit even at large widths.
problem Understanding the convergence of attention mechanisms to the kernel regime.
method Analyzes linearized attention and its relationship to the NTK limit, considering practical widths and conditions.
result Linearized attention does not converge to its NTK limit at any practical width, revealing a fundamental trade-off.
Uniform bounds for neural networks' generalization error in overparameterized settings.
problem Generalization error in overparameterized neural networks.
method Neural Tangent kernel theory and Mercer decomposition of the NT kernel in spherical harmonics.
result Uniform generalization bounds for overparameterized neural networks in RKHS.
State-of-the-art neural networks are heavily over-parameterized, making the optimization algorithm a crucial ingredient for learning predictive models with good generalization properties. A recent line of work has shown that in a certain over-parameterized regime, the learning dynamics of gradient descent are governed …
This paper shows universality in spectrum behavior for random inner-product kernel matrices in polynomial regime.
problem Understanding spectrum behavior of random inner-product kernel matrices in polynomial regime.
method Analyzing matrices formed by a nonlinear function applied entrywise to a sample-covariance matrix, considering i.i.d. entries with all finite moments.
result The spectrum of random inner-product kernel matrices is universally described by the free convolution of the semicircular and Marčenko-Pastur distributions, with relative weights given by expanding the nonlinear function in the Hermite basis.
We analyze kernel matrices in polynomial high-dimensional settings and explain double descent in KRR.
problem Understanding the spectrum of kernel matrices in polynomial high-dimensional settings and its implications for KRR risk.
method Generalized decomposition of kernel matrices into low-rank spike matrix, identity, and Gegenbauer matrix.
result The test error in KRR can exhibit double descent behavior, depending on effective regularization and signal-to-noise ratio.
Deep Gaussian Processes with polynomial kernels can collapse rapidly without proper hyperparameter tuning.
problem The collapse of Deep Gaussian Processes with polynomial kernels without careful hyperparameter tuning.
method Analysis using the Berry-Esseen Theorem and observation of prior behavior.
result The prior of a Deep Gaussian Process collapses rapidly towards zero or places negligible mass on low norm functions without proper hyperparameter tuning.
We derive finite width and depth corrections for the Neural Tangent Kernel (NTK) of ResNets and DenseNets. Our analysis reveals that finite size residual architectures are initialized much closer to the "kernel regime" than their vanilla counterparts: while in networks that do not use skip connections, convergence to t…
Study on KRR with power-law data, showing better sample complexity.
problem High-dimensional kernel ridge regression with anisotropic power-law covariance.
method Explicit characterization of kernel spectrum and asymptotic analysis of excess risk.
result Sample complexity is governed by effective dimension, not ambient dimension.
Neural networks can learn useful representations that kernels can't.
problem Learning functions that depend on only a few relevant directions.
method Gradient descent on a two-layer neural network.
result Improved sample complexity for learning polynomials.
New framework establishes positivity of DNTK for PINNs.
problem Establishing positivity of NTK for PINNs with multiple differential operators.
method Proposed Differential Neural Tangent Kernel (DNTK) for PINNs.
result Positivity of infinite width DNTK for various activation functions and differential operators.
This work shows linear convergence for two-layer neural networks in mean-field regime.
problem Optimizing two-layer neural networks in the mean-field regime.
method Mean-field analysis and continuous-time noisy gradient descent.
result Establishes linear convergence rate for two-layer neural networks.
Neural networks can learn kernel machines with a data-dependent kernel.
problem Can neural networks in the rich feature learning regime learn a kernel machine?
method Demonstrated silent alignment effect in neural networks, showing they can learn a kernel machine with a data-dependent kernel.
result Neural networks in the rich feature learning regime can learn a kernel machine with a data-dependent kernel due to silent alignment.
New insights into how neural networks learn features, especially when they are very wide.
problem Understanding how gradient flow in wide neural networks selects solutions, especially in the feature-learning regime.
method Axiomatizing the canonical regularizer as a function-space energy and lift, and deriving geodesic ridge for the feature-learning regime.
result Gradient flow in feature-learning networks biases towards ridge regularization, distorting the inductive bias and damaging pretrained networks.
Generative models use kernel smoothing for conditioning on small example sets.
problem Improving generative models' performance with limited conditioning examples.
method Showed that cross-attention conditioning is equivalent to kernel smoothing, specifically a Nadaraya--Watson kernel smoother.
result The approach predicts and confirms three failure regimes for kernel-based conditioning.
In this paper, we theoretically prove that the deep ReLU neural networks do not lie in spurious local minima in the loss landscape under the Neural Tangent Kernel (NTK) regime, that is, in the gradient descent training dynamics of the deep ReLU neural networks whose parameters are initialized by a normal distribution i…