New method designs joint initial noises for diffusion models to improve diversity and alignment.
problem Independent initial noises limit diversity in generated images.
method Coupling of initial noises, maintaining Gaussian distribution while allowing dependence.
result Repulsive Gaussian coupling improves diversity without increasing sampling cost.
Improves state space models' resistance to noise.
problem State space models' initialization assumes noise-free data, which is often violated.
method Uncertainty-aware initialization for state space models, reformulating HiPPO with measurement noise.
result Improves model resistance to noise at training and inference time.
FCNv2 robustness tested under noise and random initial conditions.
problem Assessing AI weather forecasting model robustness to input noise.
method Two experiments with varying noise levels and random initial conditions.
result FCNv2 preserves hurricane features under low to moderate noise, but underestimates intensity and persistence.
Wave maps with noise can lead to self-similar blowup from arbitrary initial data.
problem Analyzing self-similar blowup in wave maps with additive noise.
method Stochastic perturbation of wave maps in supercritical dimensions.
result Self-similar blowup with positive probability for arbitrary corotational initial data.
New insights show stochastic initialization prevents token clustering in deep Transformers.
problem Understanding token dynamics in deep stochastic Transformers.
method Analysis of deep Transformers with random initialization noise, proving convergence to an interacting-particle system on the sphere.
result Initialization noise prevents token clustering, leading to antipodal formations.
GD with large init shows incremental learning in matrix factorization.
problem Understanding GD's behavior with large initial values in matrix factorization.
method Signal-to-noise ratio concepts and inductive arguments.
result Uncovering an incremental learning phenomenon in GD with large initialization.
Standard practice in training neural networks involves initializing the weights in an independent fashion. The results of recent work suggest that feature "diversity" at initialization plays an important role in training the network. However, other initialization schemes with reduced feature diversity have also been sh…
Combines PCA and AMP for better signal estimation in noisy data.
problem Estimating a rank-1 signal in rotationally invariant noise.
method Combines PCA and AMP, with PCA initialization at the start of AMP.
result Rigorous asymptotic characterization of the new estimator's performance.
Gradient descent with noise converges to a unique optimum in nonconvex matrix factorization.
problem Gradient descent with noise converges to a unique optimum in nonconvex matrix factorization.
method A perturbed form of gradient descent with arbitrary initialization.
result Gradient descent with noise converges to a unique optimum.
Efficient method for tensor linear form inference with noisy incomplete data.
problem Statistical inference of tensor linear forms with incomplete and noisy observations.
method Initial estimate + debiasing + one-step power iteration.
result Optimal uncertainty quantification and statistical-to-computational gaps examined.
Noise stabilizes solutions to transport equations, preventing blow-up.
problem Proving global existence and uniqueness of solutions to stochastic transport equations.
method Characteristics-based techniques exploiting the geometric structure of transport equations.
result Noise prevents blow-up in deterministic solutions and ensures global existence and uniqueness of solutions.
Adversarial noises are linearly separable for random neural networks.
problem The challenge of adversarial examples in neural networks.
method Theoretical proof and empirical evidence for two-layer networks with random initialization and neural tangent kernel setup.
result Adversarial noises are linearly separable with corresponding labels.
Re-initializing neural networks improves generalization but not as much as other techniques.
problem Understanding when and how re-initialization improves neural network performance.
method Empirical comparison of re-initialization with standard training and various regularization techniques.
result Re-initialization is beneficial for generalization but not as much as other techniques, especially when combined with careful tuning of hyperparameters.
SGD transitions between maxima and minima with varying time scales.
problem Understanding SGD's behavior near critical points in noisy landscapes.
method Analyzing SGD convergence and escape dynamics in 1D landscapes with infinite- and finite-variance noise.
result SGD reliably moves to the basin's minimum unless close to a local maximum, where it can linger.
Study efficient power iteration for tensor models, proving convergence under specific conditions.
problem Simultaneous alternating power iteration for fixed-order asymmetric rank-one spiked tensor models.
method Finite-iteration local theory, geometrically decaying transient, fixed-order multilinear noise event, warm-start mechanism.
result Convergence to the unique informative local fixed point under specific conditions.
Paper proposes Sp-GD for sparse max-affine regression with theoretical guarantees.
problem Sparse max-affine regression model selection and estimation.
method Sparse Gradient Descent (Sp-GD) initialization using sparse PCA and covering search.
result Sp-GD provides ε-accurate estimates with optimal number of observations.
Study spectral learning for odeco tensors, addressing initialization bottlenecks.
problem Recovering orthogonally decomposable tensors under noise.
method Investigates perturbation bounds, non-convex optimization, and initialization strategies.
result Initialization is the main bottleneck for efficient algorithms.
Confidence-based filtering reveals latent structure in diffusion models.
problem Unclear latent structure in diffusion models.
method Confidence scores from a classifier.
result Class-relevant latent structure emerges under confidence-based filtering.
In label-noise learning, \textit{noise transition matrix}, denoting the probabilities that clean labels flip into noisy labels, plays a central role in building \textit{statistically consistent classifiers}. Existing theories have shown that the transition matrix can be learned by exploiting \textit{anchor points} (i.e…
Noise in SGD affects overparameterized models, favoring sparse solutions.
problem Understanding and mitigating implicit bias in SGD with parameter-dependent noise.
method Theoretical analysis of a quadratically-parameterized model with label noise and Gaussian noise.
result SGD with label noise recovers sparse ground-truth solutions, while SGD with Gaussian noise overfits dense solutions.
Study well-posedness of SPDE on Riemannian manifolds with rough initial conditions.
problem Well-posedness of parabolic Anderson model on Riemannian manifolds with rough initial conditions.
method Construct intrinsic Gaussian noises, explore global geometry, use Feynman-Kac formula.
result Show well-posedness with non-positive curvature and conditions on α. We solve image inverse problems using a flow-based noise model.
problem Image inverse problems with complex noise patterns.
method Normalizing flow prior for maximum a posteriori estimation.
result Empirical validation on various inverse problems.
Study examines noise sensitivity of DNNs for binary classification.
problem Understanding non-robustness of DNN classifiers under noise.
method Defined and extended noise sensitivity and stability concepts for Boolean functions, applied to DNN models.
result Sorted out the relation between definitions and properties of DNN architectures under noise.
Two-layer CNNs can overfit well if initialized correctly.
problem Understanding the conditions for benign overfitting in over-parameterized CNNs.
method Extending analysis to fully trainable two-layer CNNs, examining initialization scaling effects.
result Initialization scaling of the output layer is crucial; large scales lead to fixed output behavior, small scales to complex interactions.
Modern neural networks are typically trained in an over-parameterized regime where the parameters of the model far exceed the size of the training data. Such neural networks in principle have the capacity to (over)fit any set of labels including pure noise. Despite this, somewhat paradoxically, neural network models tr…
In this paper, we study the minimax optimization problem in the smooth and strongly convex-strongly concave setting when we have access to noisy estimates of gradients. In particular, we first analyze the stochastic Gradient Descent Ascent (GDA) method with constant stepsize, and show that it converges to a neighborhoo…
Because large, human-annotated datasets suffer from labeling errors, it is crucial to be able to train deep neural networks in the presence of label noise. While training image classification models with label noise have received much attention, training text classification models have not. In this paper, we propose an…
Method estimates noise variance in Gaussian process regression.
problem Estimating noise variance in Gaussian process regression models.
method Reduces hyperparameter space, uses marginal likelihood function, derives bounds and asymptotes.
result Computational advantages and robustness compared to traditional methods.
Neural networks learn clean data patterns first, then noisy data, leading to improved performance initially but deteriorating later.
problem Improvement in prediction error on clean data during early training of neural networks with noisy labels.
method Theoretical analysis and experiments to explore the dynamics of gradient descent and the impact of clean and noisy data.
result Neural networks prioritize learning clean data patterns first, leading to improved performance initially but deteriorating later due to diminishing gradient dominance of clean samples over noisy ones.
The paper explores how symmetries and noise in SGD influence parameter dynamics.
problem Understanding the dynamics of parameter updates in SGD with symmetries.
method Proved the existence of noise equilibria and showed their role in balancing gradient noise.
result Gradient noise creates a systematic motion of parameters to a unique fixed point, called noise equilibria.
Numerous empirical evidence has corroborated that the noise plays a crucial rule in effective and efficient training of neural networks. The theory behind, however, is still largely unknown. This paper studies this fundamental problem through training a simple two-layer convolutional neural network model. Although trai…
We introduce a training method for both better word representation and performance, which we call GROVER (Gradual Rumination On the Vector with maskERs). The method is to gradually and iteratively add random noises to word embeddings while training a model. GROVER first starts from conventional training process, and th…
SGD-trained neural networks generalize well even with adversarial label noise.
problem Generalization of neural networks trained on adversarial label noise.
method Training a one-hidden-layer neural network with SGD on arbitrary width networks.
result SGD-trained networks achieve classification accuracy competitive with the best halfspace over adversarial label noise.
Algorithm recovers factors of rank-1 matrices from noisy measurements.
problem Estimating factors of a rank-1 matrix from nonlinearly transformed and noisy measurements.
method Alternating minimization with random initialization and analysis of empirical error recursion.
result Algorithm converges geometrically fast from random initialization, with sharp guarantees.
The paper identifies generators of linear SDEs with noise types.
problem Identifying the generator of linear SDEs from their solution distribution.
method Deriving sufficient and necessary conditions for additive noise, and sufficient conditions for multiplicative noise.
result Generic conditions for identifying the generator of linear SDEs with both types of noise.
Noise Injection Node Regularization improves robustness in neural networks.
problem Improving robustness of neural networks against various perturbations.
method Injecting structured noise into neural networks during training.
result Significant improvement in robustness against data perturbations.
Antithetic noise improves diffusion models' uncertainty quantification.
problem Improving uncertainty quantification in diffusion models.
method Pairing each noise sample with its negation, leading to strong negative correlation.
result Substantially more reliable uncertainty quantification with up to 90% narrower confidence intervals.
Small initialization improves tensor recovery from noisy data.
problem Recovering low-tubal-rank tensors from noisy measurements.
method Factorized gradient descent with small initialization.
result Achieves nearly minimax optimal recovery error.
Data-driven discovery of differential equations has been an emerging research topic. We propose a novel algorithm subsampling-based threshold sparse Bayesian regression (SubTSBR) to tackle high noise and outliers. The subsampling technique is used for improving the accuracy of the Bayesian learning algorithm. It has tw…
Paper confirms winning tickets have sharp minima, useful for generalization.
problem Explaining why over-parameterized models generalize well despite having sharp minima.
method PAC-Bayesian theory applied to analyze winning tickets and their generalization behavior.
result PAC-Bayesian theory confirms winning tickets have relatively sharp minima, a disadvantage for generalization.
New method finds optimal training stop point with noisy labeled data.
problem Finding optimal training stop point with noisy labeled data.
method Analyzed training accuracy rate changes for different noise ratios to identify a training stop region. Developed a heuristic algorithm based on a small-learning assumption.
result Identified optimal training stop point at or close to maximum obtainable test accuracy.
Zeroth-order optimization methods lack inherent privacy guarantees.
problem Ensuring differential privacy in zeroth-order optimization methods.
method Analyzing ZO-GD with and without random initialization for convex and strongly convex objectives.
result ZO-GD is not differentially private for strongly convex objectives and can have superlinear privacy loss.
Multi-view subspace learning (MSL) aims to find a low-dimensional subspace of the data obtained from multiple views. Different from single view case, MSL should take both common and specific knowledge among different views into consideration. To enhance the robustness of model, the complexity, non-consistency and simil…
Scale-free distributions and correlation functions found in financial data are reminiscent of the scale invariance of physical observables in the vicinity of a critical point. Here, we present empirical evidence for a transition phenomenon, accompanied by a symmetry breaking, in the investors' demand for stocks. We stu…
VFMs use noise adapters to conditionally generate images in one step.
problem Conditional image generation with iterative models is slow and requires explicit sampling paths.
method Developed a variational flow map framework that learns noise distributions for conditional sampling.
result VFMs achieve well-calibrated conditional samples in a single forward pass.
Study improves parameter estimation for SDEs driven by Levy noise.
problem Challenges in estimating parameters of SDEs with non-Gaussian noises.
method Introduces PEnet, a CNN-LSTM model for efficient parameter estimation.
result PEnet offers superior accuracy and adaptability for various SDE scenarios.
Image denoising techniques are essential to reducing noise levels and enhancing diagnosis reliability in low-dose computed tomography (CT). Machine learning based denoising methods have shown great potential in removing the complex and spatial-variant noises in CT images. However, some residue artifacts would appear in…
Sub-gradient method recovers low-rank matrices robustly from noisy measurements.
problem Recovering low-rank matrices from noisy measurements with unknown rank.
method Sub-gradient method with small initialization, robust to over-parameterization and noise.
result Sub-gradient method converges exponentially fast to the true solution under noisy and over-parameterized conditions.