Random ReLU features are shown to be a universally consistent learning algorithm but struggle with complex functions.
problem Approximating complex functions with random ReLU features.
method Study of random ReLU features through their RKHS and composition of functions.
result Random ReLU features can efficiently approximate complex functions but not as well as multi-layer ReLU networks.
Gradient descent amplifies random features in neural networks to useful ones.
problem Generalization in neural networks trained on corrupted data.
method Characterization of feature-learning process in two-layer ReLU networks trained by gradient descent.
result Gradient descent amplifies random features to useful ones, achieving near optimal generalization error.
Our paper characterizes how ReLU affects GD's implicit bias in high-dimensional neural networks.
problem Understanding the implicit bias of gradient descent on neural networks.
method Novel primal-dual analysis tracking predictions and coefficients.
result The implicit bias approximates the minimum- ℓ 2 \ell_2 ℓ 2 -norm solution with high probability. Random features can't explain neural networks' success.
problem Understanding neural networks' empirical success.
method Analyzing random features' limitations on learning neural networks.
result Random features can't learn even a single ReLU neuron with standard Gaussian inputs.
Efficiently applies NTK to large-scale datasets using random features.
problem Computational limitations of kernel methods for large-scale datasets.
method Proposes a sketching-based algorithm combining random features of arc-cosine kernels to construct an efficient feature map of the NTK.
result Achieves comparable error bounds to exact kernel methods but with significantly reduced feature dimensionality.
Optimized nonlinearities enhance generalization in random feature models.
problem Improving generalization performance in random feature models.
method Analyzed and defined optimal nonlinearities from Gaussian model parameters.
result Optimized nonlinearities achieve better generalization performance than ReLU.
Neural networks can overfit perfectly to noisy data and then grok near-optimal generalization.
problem Neural networks' ability to overfit perfectly to noisy data and then generalize near-optimally.
method Two-layer ReLU networks trained by gradient descent on XOR cluster data.
result Neural networks can achieve perfect fit to noisy training data and then grok near-optimal generalization.
We propose semi-random features for nonlinear function approximation. The flexibility of semi-random feature lies between the fully adjustable units in deep learning and the random features used in kernel methods. For one hidden layer models with semi-random features, we prove with no unrealistic assumptions that the m…
The study examines Fisher information matrices and neural tangent kernels for simple ReLU networks with random weights.
problem Understanding the relationship between Fisher information matrices and neural tangent kernels for 2-layer ReLU networks.
method Analyzes Fisher information matrices and neural tangent kernels for 2-layer ReLU networks with random hidden weights, focusing on spectral decomposition and eigenfunctions.
result Obtained an approximation formula for functions represented by 2-layer neural networks.
We consider a neural network architecture with randomized features, a sign-splitter, followed by rectified linear units (ReLU). We prove that our architecture exhibits robustness to the input perturbation: the output feature of the neural network exhibits a Lipschitz continuity in terms of the input perturbation. We fu…
Bayesian ReLU nets fix asymptotic overconfidence with infinite features.
problem Bayesian ReLU nets can be asymptotically overconfident far from training data.
method Extend finite ReLU BNNs with infinite ReLU features via a Gaussian process.
result The resulting model is asymptotically maximally uncertain far from the data.
SGD on wide neural networks generalizes well, with bounds independent of width.
problem Generalizing from training data to unseen data in wide neural networks.
method Stochastic Gradient Descent (SGD) on over-parameterized ReLU networks.
result Generalization error bound of i l d e O ( n − 1 / 2 ) ilde{\mathcal{O}}(n^{-1/2}) i l d e O ( n − 1/2 ) independent of network width. The paper proves deep ReLU networks can die and proposes a new initialization method to prevent it.
problem Dying ReLU neurons in deep neural networks.
method The paper rigorously proves the dying ReLU problem and proposes a new randomized asymmetric initialization method.
result The new initialization method effectively prevents the dying ReLU problem.
Framework explains how dual deep networks learn features from unlabeled data.
problem Understanding self-supervised learning with dual deep networks.
method Theoretical framework and hierarchical latent tree model.
result Deep ReLU networks learn latent variables through contrastive SSL.
Most random ReLU networks are vulnerable to small, Euclidean adversarial perturbations.
problem Vulnerability of ReLU networks to adversarial attacks.
method Analysis of random ReLU networks with decreasing dimensions, using gradient flow and descent.
result Most examples can be perturbed by small Euclidean distances via gradient methods.
Attention layers are sensitive to single words, improving generalization over random features.
problem Understanding why attention layers are effective in NLP tasks.
method Study of word sensitivity in random features using BERT-Base word embeddings.
result Attention layers have high word sensitivity, improving generalization over random features.
New bounds on ReLU networks for low-regular functions.
problem Bounding approximation error for ReLU networks on low-regular functions.
method Complexity analysis of Fourier features residual networks to ReLU networks.
result Approximation error bound proportional to target function norm and inversely proportional to network width and depth.
Randomly initialized ReLU networks of depth two can approximate smooth functions well.
problem Approximation power of two-layer networks of random ReLUs.
method Harmonic analysis and ridgelet representation theory for upper bounds, dimensionality arguments for lower bounds.
result Near-matching upper and lower bounds for L 2 L_2 L 2 -approximation and Sobolev norms. This paper characterizes how randomized neural networks generalize well in multi-dimensional tasks.
problem Understanding the generalization of randomized neural networks in multi-dimensional tasks.
method Characterizes RSNs as an IGAM formalized by an optimization problem with a regularization functional and loss.
result RSNs generalize well in multi-dimensional tasks, akin to spline regression under certain conditions.
The paper shows how random ReLU networks converge to smooth splines.
problem Understanding the behavior of shallow ReLU neural networks with random weights.
method Mathematical analysis of L2-regularized regression and gradient descent.
result Random ReLU networks converge to smooth splines as the number of hidden nodes increases.
Neural networks can separate non-separable data using feature maps.
problem Non-separable data in neural networks.
method Characterization of feedforward neural networks and use of feature maps.
result ReLU neural networks can separate concentric data.
The paper examines topological features of ReLU networks and their relation to decision boundaries and training loss.
problem Understanding the topological structure of ReLU neural network activation patterns.
method Polytope decomposition of feature space, Fiedler partition of dual graph, homology computation of cellular decomposition.
result The Fiedler partition of the dual graph correlates with decision boundaries in binary classification tasks, and similar patterns in training loss and polyhedral cell-count emerge in regression tasks.
The paper analyzes the role of ReLU gates in deep learning networks.
problem Understanding the role of gates in deep learning networks.
method Developed neural path features (NPF) and neural path values (NPV) to characterize the active sub-networks during training.
result The neural path kernel associated with NPFs is a fundamental quantity that characterizes the information stored in the gates of a DNN.
Two-layer ReLU networks outperform kernel methods in teacher-student settings.
problem Understanding the excess risk of two-layer ReLU neural networks in teacher-student models.
method Investigated a two-phase training process for a student network, comparing it to kernel methods.
result The student network reaches near-global optimality and outperforms kernel methods in minimax optimal rate.
Echo state networks with random weights can approximate any continuous system.
problem Approximating continuous dynamical systems using echo state networks.
method Randomly generated internal weights and a sampling procedure for activation functions.
result Echo state networks with random weights can approximate any continuous casual time-invariant operators with high probability.
Gradient descent optimizes deep ReLU networks with proper initialization.
problem Training deep neural networks with ReLU activation.
method Gradient descent and stochastic gradient descent with proper random weight initialization.
result Gradient descent finds global minima for over-parameterized deep ReLU networks.
Paper tackles learning ReLU networks for binary classification with linearly separable data.
problem Learning two-layer ReLU networks for binary classification with linearly separable data.
method Stochastic gradient descent (SGD) algorithm with random noise perturbation.
result Proves global optimality of SGD for training any single-hidden-layer ReLU network.
We find that CNNs with BN and ReLU exhibit feature sparsity through selective feature penalization.
problem Feature sparsity in CNNs trained with specific techniques.
method Empirical study and hypothesis testing on sparsification mechanisms.
result Selective feature penalization leads to feature sparsity in CNNs, comparable to explicit pruning.
Gradient descent achieves good generalization for over-parameterized deep ReLU networks.
problem Understanding good generalization in over-parameterized deep neural networks.
method Algorithm-dependent generalization error bound for deep ReLU networks using gradient descent.
result Gradient descent with proper initialization can achieve arbitrarily small generalization error for over-parameterized DNNs.
The paper improves lower bounds on deep neural network trajectories.
problem Understanding the expressivity of deep neural networks.
method Generalized method for lower bounding trajectory growth in random sparse deep ReLU networks.
result Trajectory growth can remain exponential in depth with sparse variants of random nets.
Gradient descent learns ReLU functions with non-zero bias efficiently.
problem Learning ReLU functions with non-zero bias under Gaussian distributions.
method Gradient descent starting from random initialization.
result Gradient descent achieves near-optimal error with high probability.
Study on function sensitivity in random DNNs using large deviation theory.
problem Understanding function sensitivity in finite-size deep neural networks.
method Large deviation theory and path integral analysis applied to random DNNs with ReLU and sign activations.
result Random DNNs with ReLU activations are more robust to parameter perturbations.
Study of two-layer ReLU neural network phase diagram at infinite-width limit.
problem Characterize the dynamical regimes of two-layer ReLU neural networks.
method Combining experimental and theoretical approaches, including phase diagram analogy.
result Identification of three regimes: linear, critical, and condensed.
The study calculates the injectivity capacity of ReLU networks using a novel mathematical approach.
problem Determining the injectivity capacity of ReLU networks layers.
method Employing fully lifted random duality theory (fl RDT) to handle the ℓ 0 \ell_0 ℓ 0 spherical perceptron and implicitly the ReLU layers injectivity. result The lifting mechanism converges remarkably fast with relative corrections not exceeding 0.1%.
ReLU networks don't exponentially distort curve lengths as previously thought.
problem Understanding how neural networks distort curve lengths with depth.
method Analyzing expected length distortion of ReLU networks with random initialization.
result Expected length distortion does not grow with depth, and shrinks slightly.
The paper reveals how clustering in ReLU networks enhances interpretability.
problem Improving interpretability of deep learning models.
method Clustering of ReLU neuron patterns to reveal underlying affine maps.
result The network's predictions can be explained by feature importance within clusters.
The Rectified Linear Unit (ReLU) is a foundational activation function in artficial neural networks. Recent literature frequently misattributes its origin to the 2018 (initial) version of this paper, which exclusively investigated ReLU at the classification layer. This paper formally corrects the citation record by tra…
Gradient descent trains both layers of a ReLU network to fit a linear model.
problem Training dynamics of a ReLU network to fit a linear target function.
method Jointly training both layers of a one-hidden-layer ReLU network in a realizable setting with Gaussian inputs and labels.
result Gradient descent from a small random initialization converges to a global minimizer at a linear rate with optimal sample complexity.
Deep ReLU networks generalize via piecewise linear interpolation.
problem Understanding how deep neural networks generalize well on real datasets despite memorizing training samples.
method Demonstrated that deep ReLU networks generalize via piecewise linear interpolation, providing a quantified analysis of generalization intervals.
result Generalization intervals of ReLU networks behave similarly along pairwise directions between samples of the same label in both real and random cases, suggesting the same interpolation mechanism.
New bounds explain deterministic non-smooth deep nets without large Lipschitz constants.
problem Challenges in explaining generalization of deterministic non-smooth deep nets.
method De-randomized PAC-Bayes margin bounds for deterministic non-convex and non-smooth predictors.
result New bounds avoid large Lipschitz constants, providing generalization guarantees.
Study shows efficient neural network approach for stochastic bandits.
problem Optimizing decisions in uncertain environments with neural network models.
method OFU-ReLU algorithm that balances exploration and exploitation, using a transformed feature space.
result Achieves i l d e O ( T ) ilde{O}(\sqrt{T}) i l d e O ( T ) regret guarantee for stochastic bandits with ReLU neural networks. Injectivity of ReLU networks is characterized for generative models and inverse problems.
problem Injectivity in ReLU networks for generative models and inverse problems.
method Layerwise analysis, worst-case Lipschitz constants, differential topology, random projections.
result Global injectivity of ReLU networks requires expansivity between 3.4 and 10.5 for Gaussian matrices.
SGD fails to converge for deep ReLU networks with limited random initializations.
problem SGD convergence in deep neural networks with limited random initializations.
method Analysis of four discretization parameters: network architecture, training data, gradient steps, and random initializations.
result SGD fails to converge for ReLU networks with depth much larger than width.
Deep and wide ReLU networks learn data-dependent features even in the lazy training regime.
problem Understanding the behavior of neural networks with finite depth and width.
method Analyzing the mean and variance of the neural tangent kernel (NTK) in a randomly initialized ReLU network.
result The NTK has a non-trivial evolution during training, with the mean of its first SGD update being exponential in the ratio of depth to width.
The angular power spectrum characterizes neural network complexity.
problem Characterizing the complexity of deep neural networks.
method Using the angular power spectrum of the limiting field to characterize network complexity.
result Classified neural networks as low-disorder, sparse, or high-disorder.
Exact bounds derived for neural network outputs with noisy inputs.
problem Bounding the output distribution of neural networks with random inputs.
method Applying ReLU NNs to derive bounds for general NNs, then using these to find exact error guarantees.
result Exact upper and lower bounds for the output distribution of neural networks with random inputs.
In this paper we investigate the performance of different types of rectified activation functions in convolutional neural network: standard rectified linear unit (ReLU), leaky rectified linear unit (Leaky ReLU), parametric rectified linear unit (PReLU) and a new randomized leaky rectified linear units (RReLU). We evalu…
This paper extends neural collapse to class-imbalanced datasets using an unconstrained ReLU feature model.
problem Understanding neural collapse in class-imbalanced datasets with cross-entropy loss.
method Generalized neural collapse to class-imbalanced settings using an unconstrained ReLU feature model.
result Class-means converge to orthogonal vectors with different lengths, and classifier weights align to these vectors.