This paper presents a margin-based multiclass generalization bound for neural networks that scales with their margin-normalized "spectral complexity": their Lipschitz constant, meaning the product of the spectral norms of the weight matrices, times a certain correction factor. This bound is empirically investigated for…
On-device research index
arXiv research
A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.
169,051 papers · 148 categories
Trend · papers per month
48 results for “spectrally normalized margins”
Improves DNN robustness to adversarial attacks.
problem Adversarial attacks degrade DNN robustness.
method Spectral normalization of weight matrices.
result Spectral normalization enhances DNN generalization.
This paper resolves Breiman's dilemma in neural networks by analyzing phase transitions of margin dynamics.
problem Breiman's dilemma in neural networks: uniform margin improvement does not guarantee reduced generalization errors.
method Revisiting Breiman's dilemma in deep neural networks with spectrally normalized margins, analyzing phase transitions of normalized margin distributions.
result Margin-based generalization bounds can predict test error trends during training phase transitions.
Paper improves deep neural networks' generalization by focusing on margin distribution complexity.
problem Improving deep neural networks' generalization performance.
method Proves a generalization upper bound based on margin distribution statistics and optimizes a convex margin distribution loss function.
result Optimizing the ratio of margin standard deviation to expected margin enhances generalization performance.
The paper proves geometric and spectral alignment for deep neural networks.
problem Understanding the singular spectra of deep neural network layers.
method Proves deterministic quotient-geometric estimates for singular spectra of Frobenius-normalized layer factors.
result Exact power-law spectra form a trace-normalized Cartan orbit under Frobenius normalization.
ACFS optimizes spectral risk under decision-dependent uncertainty using adaptive forest sampling.
problem Minimizing spectral risk with decision-dependent uncertainty.
method ACFS integrates Generalised Random Forests, CEM-guided exploration, rank-weighted augmentation, and multi-start refinement.
result ACFS achieves lowest median oracle spectral risk on both benchmarks.
Study connects spectral clustering to maximum margin and level set estimation.
problem Connecting spectral clustering to maximum margin and level set estimation.
method Obtained bounds on eigenvectors of graph Laplacian matrices in terms of cluster separation and connectivity. Showed sensitivity mitigation by removing outliers and estimating level sets.
result Spectral clustering converges to maximum margin clustering as scaling parameter approaches zero.
Spectral algorithms improve under covariate shift with novel weighted techniques.
problem Improving spectral algorithms' performance under covariate shift.
method Analysis of spectral algorithms in non-parametric regression over RKHS, proposing a weighted spectral algorithm with clipped weights.
result Normalized weighted spectral algorithm achieves optimal capacity-independent convergence rates, and clipped weights can approach optimal capacity-dependent rates.
Algorithm learns dynamics from past observations.
problem Learning a nonlinear dynamical system.
method Spectral filtering, online convex optimization.
result Vanishing prediction error for marginally stable systems.
Batch normalization biases linear models towards uniform margins, improving performance in binary classification.
problem Understanding the implicit bias of batch normalization in linear models and neural networks.
method Analyzing gradient descent convergence on linear models and two-layer CNNs with batch normalization.
result Gradient descent with batch normalization in linear models converges to a uniform margin classifier with an exponential convergence rate.
The generalization error of deep neural networks via their classification margin is studied in this work. Our approach is based on the Jacobian matrix of a deep neural network and can be applied to networks with arbitrary non-linearities and pooling layers, and to networks with different architectures such as feed forw…
One of the challenges in the study of generative adversarial networks is the instability of its training. In this paper, we propose a novel weight normalization technique called spectral normalization to stabilize the training of the discriminator. Our new normalization technique is computationally light and easy to in…
Optimizes risk measures given known marginal distributions of two unknown factors.
problem Determining an upper bound for spectral risk measures with unknown joint distribution.
method Introduces Maximum Spectral Measure (MSP) as a worst-case risk measure, formulated as an optimization problem with a more general objective function.
result Characterizes the continuity properties of the optimal value function and optimal solution set with respect to marginal distributions.
Kernel estimator improves spectral risk measure estimation.
problem Estimating spectral risk measures accurately.
method Kernel-based estimation of L-statistics for SRMs.
result Kernel estimator is strongly consistent and asymptotically normal.
A new method speeds up spectral normalization for neural nets.
problem Efficiently controlling the spectral norm of convolutional layers.
method Depthwise separable convolutions with spectral normalization.
result Significant reduction in computational and memory costs.
The late-stage training dynamics of (stochastic) subgradient descent on homogeneous neural networkscs.LG
SGD converges to critical points of normalized margin in late-stage training for homogeneous neural networks.
problem Analyzing the implicit bias of SGD on homogeneous neural networks.
method Interpreting SGD dynamics as an Euler-like discretization of a conservative field flow associated with the normalized classification margin.
result Normalized SGD iterates converge to the set of critical points of the normalized margin at late-stage training.
We study the problem of determining the optimal low dimensional projection for maximising the separability of a binary partition of an unlabelled dataset, as measured by spectral graph theory. This is achieved by finding projections which minimise the second eigenvalue of the graph Laplacian of the projected data, whic…
In this paper, we consider unsupervised partitioning problems, such as clustering, image segmentation, video segmentation and other change-point detection problems. We focus on partitioning problems based explicitly or implicitly on the minimization of Euclidean distortions, which include mean-based change-point detect…
Spectral normalization stabilizes GANs by controlling gradient explosion and vanishing.
problem Stability and sample quality issues in GAN training.
method Spectral normalization controls gradient explosion and vanishing, improving GAN training stability and sample quality.
result Bidirectional Scaled Spectral Normalization (BSSN) outperforms standard spectral normalization in sample quality and training stability.
Spectral clustering is a technique that clusters elements using the top few eigenvectors of their (possibly normalized) similarity matrix. The quality of spectral clustering is closely tied to the convergence properties of these principal eigenvectors. This rate of convergence has been shown to be identical for both th…
New method improves GAN training stability and quality.
problem Improving training stability and sample quality in GANs.
method Proposes a new method for Lipschitz continuity in GANs that is efficient and unbiased.
result Demonstrates the effectiveness of the new method in various GAN training scenarios.
The multiplier spectral curve of a conformal torus in the 4-sphere is essentially, see arXiv:0712.2311, given by all Darboux transforms of the conformal torus. In the particular case when the conformal immersion is a Hamiltonian stationary torus in Euclidean 4-space, the left normal of the immersion is harmonic, hence …
We prove a mapping between dual and primal factor graph marginals for efficient estimation.
problem Efficient estimation of marginal densities in factor graphs.
method Local mappings derived from Fourier transforms of local factors, applied to Ising and Potts models.
result Marginal densities can be more accurately estimated in the dual domain.
Muon dynamics study uses spectral Wasserstein flow for optimization stability.
problem Optimizing deep learning models with gradient normalization.
method Introduces Spectral Wasserstein distances for matrix flows, proving equivalence with Benamou--Brenier formulation.
result Gradient-flow interpretation of mean-field normalized training dynamics.
This paper applies the Extreme-Value (EV) Generalised Pareto distribution to the extreme tails of the return distributions for the S&P500, FT100, DAX, Hang Seng, and Nikkei225 futures contracts. It then uses tail estimators from these contracts to estimate spectral risk measures, which are coherent risk measures that r…
Transformer-based method discovers objects from images without labels.
problem Discovering objects in images without labeled data.
method Graph-based approach using self-supervised transformer features and normalized graph-cut.
result Significantly boosts performance in unsupervised object discovery.
Proposes MSN to improve DNN performance and speed.
problem Improving Deep Learning model regularization and performance.
method Empirical approach to study Spectral Normalization (SN) and Mean Spectral Normalization (MSN).
result MSN significantly improves DNN performance and speed.
Solves the Wiegold problem by showing free products of left-orderable groups have normal rank > 1.
problem Wiegold problem about groups of normal rank > 1
method Topological argument and intricate construction of left-orders
result Free products of nontrivial left-orderable groups have normal rank > 1
Paper improves normalizing flows to better capture distribution tails.
problem Difficult to learn tail behavior of distributions.
method Develops a new type of flows using flexible base distributions and data-driven linear layers.
result Improves accuracy, especially on distribution tails, and generates heavy-tailed data.
Local mappings relate dual and primal factor graphs for efficient marginal probability estimation.
problem Efficient estimation of marginal probabilities in statistical physics models.
method Local mappings based on Fourier transform of local factors, applied to Ising, Potts, and clock models.
result Local extrema of fixed points are at phase transition points, and the mapping facilitates efficient estimation.
Bayesian model improves traffic prediction with uncertainty estimates.
problem Lack of uncertainty estimates in deep-learning traffic models.
method Proposes a Bayesian recurrent neural network with spectral normalization.
result Spectral normalization improves uncertainty estimates and generalizability.
The paper improves alignment methods for deep neural networks using geometric and spectral analysis.
problem Improving alignment methods for deep neural networks.
method Geometric and spectral analysis of residual Jacobian chains.
result Deterministic and margin-verified results on the transport of dominant singular subspaces across layers.
A new method estimates marginal likelihood using normalizing flows.
problem Estimating marginal likelihood in Bayesian model selection.
method Learned harmonic mean estimator using normalizing flows.
result Normalizing flows avoid the exploding variance problem.
New method controls linear systems with partial info and disturbances.
problem Controlling linear dynamical systems under partial observation and adversarial disturbances.
method Double Spectral Control (DSC) using two-level spectral approximation strategy.
result Matches best known regret guarantees with exponential runtime improvement.
Develops an oblique projection technique to approximate a foliation for non-normal dynamics.
problem Modeling dynamics far from a primary Spectral Submanifold (SSM) in non-normal systems.
method Oblique projection technique based on experimental data.
result Approximates a stable invariant foliation for non-normal dynamics efficiently.
COMET Flows model multivariate extremes with heavy tails and asymmetric dependence.
problem Normalizing flows struggle with multivariate extremes and asymmetric tail dependence.
method COMET Flows decomposes modeling into marginal and copula parts; uses tail belief and kernel density for marginals, and low-dimensional manifold for tail dependence.
result COMET Flows outperform other models in capturing heavy-tailed marginals and asymmetric tail dependence.
Generative models create paintings that match training data.
problem Creating realistic paintings using machine learning.
method Used Spectral Normalization GAN (SN-GAN) and SN-GAN with Gradient Penalty to generate paintings.
result SN-GAN produced paintings most comparable to the training dataset.
The study of spectral-tightness in Riemannian manifolds and its topological implications.
problem Understanding the spectral properties of Riemannian manifolds and their coverings.
method Analyzing the fundamental group and the Euclidean local de Rham factor to characterize spectral-tightness.
result Spectral-tightness is a topological property of the fundamental group, and it can be characterized by the dimension of the Euclidean local de Rham factor.
Spectral Risk Measures with an Application to Futures Clearinghouse Variation Margin Requirementsq-fin.RM
This paper applies an AR(1)-GARCH (1, 1) process to detail the conditional distributions of the return distributions for the S&P500, FT100, DAX, Hang Seng, and Nikkei225 futures contracts. It then uses the conditional distribution for these contracts to estimate spectral risk measures, which are coherent risk measures …
An important form of prior information in clustering comes in form of cannot-link and must-link constraints. We present a generalization of the popular spectral clustering technique which integrates such constraints. Motivated by the recently proposed -spectral clustering for the unconstrained problem, our method is…
Combines MCTM and NF for flexible multivariate density regression with interpretable marginals.
problem Difficult interpretation of flexible NF models and limitations of MCTM in flexibility.
method Hybrid approach combining MCTM for interpretable marginals and NF for complex joint distributions.
result Demonstrates versatility and improved performance compared to MCTM and other NF models.
A spacelike surface is marginally trapped if its mean curvature vector is lightlike. On any oriented spacelike surface we show that a choice of orientation of the normal bundle determines a smooth map which we call the null Gauss map of…
Principal component analysis (PCA) is arguably the most popular tool in multivariate exploratory data analysis. In this paper, we consider the question of how to handle heterogeneous variables that include continuous, binary, and ordinal. In the probabilistic interpretation of low-rank PCA, the data has a normal multiv…
We study spectral asymptotics for the Laplace operator on differential forms on a Riemannian foliated manifold equipped with a bundle-like metric in the case when the metric is blown up in directions normal to the leaves of the foliation. The asymptotical formula for the eigenvalue distribution function is obtained. Th…
Gradient descent in neural networks maximizes margin.
problem Optimizing neural networks using gradient descent.
method Gradient descent or gradient flow on homogeneous neural networks.
result Normalized margin increases over time if training loss decreases below a threshold.
We prove the existence of metrics maximizing the first eigenvalue normalized by area on closed, non-orientable surfaces assuming two spectral gap conditions. These spectral gap conditions are proved by the authors in \cite{MS3}.
Can one reduce the size of a graph without significantly altering its basic properties? The graph reduction problem is hereby approached from the perspective of restricted spectral approximation, a modification of the spectral similarity measure used for graph sparsification. This choice is motivated by the observation…
Study beta function for convex billiard maps, linking spectral invariants.
problem Understanding spectral invariants of convex billiard maps.
method Birkhoff normal form via constructive generating functions, explicit beta function formula.
result Linked spectral invariants to beta function for convex billiard maps.