Study challenges the Gaussian pre-activations assumption in neural networks.
problem Challenges the assumption that pre-activations are Gaussian in neural networks.
method Constructs pairs of activation functions and initialization distributions to ensure Gaussian pre-activations.
result Discovered constraints for ensuring Gaussian pre-activations in neural networks.
Study on MC dropout in wide neural networks and its convergence to Gaussian processes.
problem Understanding the behavior of Monte Carlo dropout in wide neural networks.
method Rigorously studied the limiting distribution of wide untrained NNs under dropout, proving convergence to Gaussian processes. Investigated correlations and non-Gaussian behavior in finite width NNs.
result Wide untrained neural networks under dropout converge to Gaussian processes for fixed sets of weights and biases.
Study infinite-depth limits of neural networks with fixed width.
problem Understanding the behavior of neural networks as depth increases with fixed width.
method Analyzing finite-width residual networks with random Gaussian weights, focusing on the infinite-depth limit.
result The pre-activations converge to a zero-drift diffusion process, differing from the infinite-width limit.
Develops EFT for ResNets, revealing limitations of kernel-only approach.
problem Limitations of kernel-only approach in deep neural networks.
method Collective kernel EFT for pre-activation ResNets based on G-only closure hierarchy. result Numerical findings show V4 equation residual accumulates to an O(1) error. Gradient descent aligns neural feature matrices with pre-activation tangent features.
problem Understanding neural feature learning mechanisms.
method Analytical proof of alignment between weight matrices and pre-activation tangent features.
result Derivative alignment occurs almost surely in high-dimensional settings.
Deep neural networks learn by averaging fast variables, revealing a Gaussian process.
problem Analyzing the complex behavior of deep neural networks (DNNs) with billions of parameters.
method Identifying slow variables that average the erratic behavior of fast microscopic variables in fully trained DNNs.
result DNN layers couple only through the second moment (kernels) of their activations and pre-activations, which fluctuate in a nearly Gaussian manner.
Theoretical limits of deep residual networks show consistent covariance structures.
problem Understanding the limits of deep residual networks.
method Analyzing the behavior of deep residual networks with skip connections as width and depth approach infinity.
result Theoretical analysis confirms that the covariance structure remains consistent regardless of the order of width and depth.
Deep gated networks help understand training and generalization in deep learning.
problem Understanding the role of SGD in training and generalization of deep neural networks with ReLU activation.
method Developed deep gated networks (DGNs) as a framework to analyze training and generalization in DNNs with ReLU activation.
result Gate adaptation is key for generalization in deep neural networks.
We describe Bayesian Layers, a module designed for fast experimentation with neural network uncertainty. It extends neural network libraries with drop-in replacements for common layers. This enables composition via a unified abstraction over deterministic and stochastic functions and allows for scalability via the unde…
New initialization schemes preserve fractional moments of weights in deep networks, improving training and test performance.
problem Heavy-tailed distribution of stochastic gradients in DNNs during training.
method Developed initialization schemes that preserve any given fractional moment of order s < 2 over layers for various activations.
result The network output admits a heavy-tailed distribution with finite moments, improving training and test performance.
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.
A number of recent papers have provided evidence that practical design questions about neural networks may be tackled theoretically by studying the behavior of random networks. However, until now the tools available for analyzing random neural networks have been relatively ad-hoc. In this work, we show that the distrib…
While the authors of Batch Normalization (BN) identify and address an important problem involved in training deep networks-- Internal Covariate Shift-- the current solution has certain drawbacks. Specifically, BN depends on batch statistics for layerwise input normalization during training which makes the estimates of …
While the authors of Batch Normalization (BN) identify and address an important problem involved in training deep networks-- \textit{Internal Covariate Shift}-- the current solution has certain drawbacks. For instance, BN depends on batch statistics for layerwise input normalization during training which makes the esti…
Bayesian Neural Networks (BNNs) have recently received increasing attention for their ability to provide well-calibrated posterior uncertainties. However, model selection---even choosing the number of nodes---remains an open question. Recent work has proposed the use of a horseshoe prior over node pre-activations of a …
Bayesian Neural Networks (BNNs) have recently received increasing attention for their ability to provide well-calibrated posterior uncertainties. However, model selection---even choosing the number of nodes---remains an open question. In this work, we apply a horseshoe prior over node pre-activations of a Bayesian neur…
A new method normalizes activations to match batch normalization without batch dependence.
problem Performance degradation with batch-independent normalization techniques.
method Proxy-Normalizing Activations
result Proxy-Normalization technique emulates batch normalization's behavior and performance.
We propose to explain the predictions of a deep neural network, by pointing to the set of what we call representer points in the training set, for a given test point prediction. Specifically, we show that we can decompose the pre-activation prediction of a neural network into a linear combination of activations of trai…
Study on learning rates in neural networks of varying depth.
problem Dependence of maximal update learning rate on network depth.
method Analysis of random fully connected ReLU networks with mean-field weight initialization.
result Maximal update learning rate scales like L−3/2 with network depth. Deep neural networks with heavy-tailed weights converge to stable distributions.
problem Understanding the convergence of heavy-tailed weights in infinitely-wide neural networks.
method Analyzing infinitely-wide multi-layer perceptrons with i.i.d. symmetric α-stable weight distributions. result The vector of pre-activation values converges to i.i.d. symmetric α-stable distributions. DeeperGCN tackles deep GCNs by overcoming vanishing gradient and over-smoothing issues.
problem Vanishing gradient, over-smoothing, and over-fitting issues in deep GCNs.
method DeeperGCN uses differentiable generalized aggregation functions and a novel normalization layer (MsgNorm) to train deep GCNs.
result DeeperGCN significantly boosts performance on large-scale graph learning tasks.
The paper analyzes various normalization methods in deep learning.
problem Lack of mathematical tools to analyze normalization methods.
method Proposed a lemma to define tools, analyzed BN, LN, WN, GN.
result Normalization methods can be unified on a sphere, improving training stability and weight norm.
FP uses random projections to train networks without feedback, achieving comparable performance to backpropagation.
problem Training neural networks without feedback from downstream layers.
method Forward Projection (FP) method that uses randomised nonlinear projections and closed-form regression.
result FP achieves comparable generalisation to backpropagation methods with a single forward pass, offering significant speedup.
Single training run learns optimal VAE parameters for various β values.
problem Training VAEs with varying β values for optimal trade-off between distortion and rate.
method Introduced Multi-Rate VAE (MR-VAE) using hypernetworks to map β to optimal parameters.
result MR-VAEs can construct the full rate-distortion curve without additional training.
New findings challenge the use of flatness measures in neural networks.
problem The validity of flatness measures in assessing generalization in neural networks.
method Analysis of Hessian-based flatness norms and their relation to generalization.
result Solutions with large weights and low loss are often sharper than expected, contradicting flatness measures.
The ability of overparameterized deep networks to generalize well has been linked to the fact that stochastic gradient descent (SGD) finds solutions that lie in flat, wide minima in the training loss -- minima where the output of the network is resilient to small random noise added to its parameters. So far this observ…
New method improves neural network verification by considering multivariate input space of ReLU neurons.
problem Improving the effectiveness of neural network verification algorithms.
method A new tightened convex relaxation for ReLU neurons considering multivariate input space.
result Our convex relaxation is significantly stronger than the commonly used univariate-input relaxation.
Spatial smoothing improves BNNs' accuracy, uncertainty, and robustness without increasing computational cost.
problem Large ensembles in BNNs increase computational cost and reduce performance.
method Spatial smoothing adds blur layers to convolutional neural networks to ensemble neighboring feature map points.
result Spatial smoothing improves BNNs' performance with fewer ensembles and enhances robustness.
Unified learning-rate scale for CNNs and ResNets, avoiding depth imbalance.
problem Challenges in choosing an appropriate learning rate for deep networks, especially as depth increases.
method Introduces Arithmetic-Mean μP (AM-μP), constraining network-wide average pre-activation second moment to a constant scale, combined with residual-aware He fan-in initialization. result Demonstrates a −3/2 scaling law for learning rates across depths, enabling zero-shot learning-rate transfer. We explore a new approach for training neural networks where all loss functions are replaced by hard constraints. The same approach is very successful in phase retrieval, where signals are reconstructed from magnitude constraints and general characteristics (sparsity, support, etc.). Instead of taking gradient steps, t…
AReLU uses attention-based rectification to improve neural network performance.
problem Improving neural network performance through better activation functions.
method Integrates attention mechanism with rectified linear unit (ReLU) to learn and scale feature maps.
result AReLU significantly boosts performance of most network architectures with minimal changes.
Gaussian processes model geospatial trajectories with uncertainty.
problem Interpolating and predicting complex spatiotemporal data.
method Gaussian process models trajectories as multidimensional Gaussian distributions.
result Gaussian processes provide a flexible and probabilistic way to interpolate geospatial data.
Independent component analysis (ICA) decomposes multivariate data into mutually independent components (ICs). The ICA model is subject to a constraint that at most one of these components is Gaussian, which is required for model identifiability. Linear non-Gaussian component analysis (LNGCA) generalizes the ICA model t…
New Gaussian min-max theorem extends classical results to non-i.i.d. Gaussian matrices.
problem Extending classical Gaussian min-max theorems to non-i.i.d. Gaussian matrices.
method Identifying a new pair of Gaussian processes that satisfy comparison inequalities.
result New Gaussian min-max and convex Gaussian min-max theorems with applications in multi-source Gaussian regression and binary classification.
GGMPs improve non-Gaussian conditional density estimation.
problem Multimodality, heteroscedasticity, and strong non-Gaussianity in conditional density estimation.
method GGMP combines local Gaussian mixture fitting, cross-input component alignment, and per-component heteroscedastic GP training.
result GGMPs improve distributional approximation on synthetic and real-world datasets.
Solves Lp-Gaussian chord Minkowski problem using Gauss curvature flow.
problem Solving the Lp-Gaussian chord Minkowski problem. method Using Gauss curvature flow to obtain smooth even solutions.
result Obtains smooth even solutions to the Lp-Gaussian chord Minkowski problem. New findings show Gaussian universality breaks down in high-dimensional linear factor mixtures.
problem The limitations of Gaussian universality in high-dimensional classification.
method Characterization of empirical risk minimization for classification under linear factor mixture models.
result Gaussian universality breaks down under high-dimensional linear factor mixtures.
Proposes efficient Gaussian approximations for non-Gaussian likelihoods.
problem Computational challenges in learning and inference with non-Gaussian likelihoods.
method Variational inference and moment matching in transformed bases.
result Good approximation quality for binary and multiclass classification.
Gaussian-SVGD dynamics converge to Gaussian distributions under certain conditions.
problem Understanding the theoretical properties of SVGD, especially for Gaussian targets.
method Detailed theoretical study of Gaussian-SVGD dynamics, considering both mean-field PDE and discrete particle systems.
result Gaussian-SVGD dynamics converge linearly to the Gaussian distribution closest to the target in KL divergence.
Sharp Gaussian isoperimetry proven along Ricci flow.
problem Proving sharp Gaussian isoperimetric inequality for Ricci flow.
method Using monotonicity formula to prove inequality.
result Exact Gaussian enlargement theorem and concentration estimates.
Gaussianization flows transform any random vector into a Gaussian, enabling efficient computation and sample generation.
problem Transforming any random vector into a Gaussian for efficient computation and sample generation.
method Iterative Gaussianization and normalizing flow model.
result Gaussianization flows are universal approximators and achieve better performance on tabular datasets.
Ens-CGP synthesizes ensemble-based inference with Gaussian processes.
problem Ensemble-based inference and Gaussian process modeling.
method Formulates Ens-CGP as a conditional Gaussian process for ensemble moments.
result Ens-CGP provides a unified probabilistic foundation for Kalman-type methods.
Geometric Gaussian approximations capture any distribution.
problem Approximating complex probability distributions.
method Geometric Gaussian approximations through diffeomorphisms or exponential maps.
result Geometric Gaussian approximations are universal, capturing any distribution.
Gaussian processes struggle with compositional functions, but deep Gaussian processes can outperform.
problem Gaussian process regression struggles with compositional functions.
method We study information-theoretic lower bounds for posterior contraction rates in Gaussian process regression for a continuous regression model.
result Posterior based on any mean-zero Gaussian process can only recover the truth at a rate strictly slower than the minimax rate for generalized additive functions.
New sparse Gaussian process method tackles unconstrained regression problems.
problem Dealing with physical systems that satisfy inequality constraints.
method Extends constrained Gaussian process by redefining hat basis functions.
result Reduces computational complexity from O(n3) to O(nm2). Wide deep neural networks with Gaussian weights approximate Gaussian processes closely.
problem Understanding the approximation of deep neural networks with Gaussian weights to Gaussian processes.
method Established novel rates for the Gaussian approximation of random deep neural networks with Gaussian parameters and Lipschitz activation functions in the wide limit.
result The distance between the network output and the Gaussian approximation scales inversely with the width of the network.
This paper proposes a method to approximate non-Gaussian likelihoods in Gaussian Processes.
problem Approximating non-Gaussian likelihoods in Gaussian Processes.
method Proposes a piece-wise constant approximation for the inverse-link function.
result Yields a closed form solution for the SVGP lower bound.
The Kalman filter is extensively used for state estimation for linear systems under Gaussian noise. When non-Gaussian Lévy noise is present, the conventional Kalman filter may fail to be effective due to the fact that the non-Gaussian Lévy noise may have infinite variance. A modified Kalman filter for linear systems wi…