Improved deep neural network generalization through noise resilience.
problem Understanding and predicting generalization error of deep neural networks.
method Noise resilience measures to predict generalization error.
result Secured 5th position in the PGDL competition at NeurIPS 2020.
Deep neural networks (DNNs) have been widely used in the fields such as natural language processing, computer vision and image recognition. But several studies have been shown that deep neural networks can be easily fooled by artificial examples with some perturbations, which are widely known as adversarial examples. A…
Paper tackles noisy neural networks and proposes a method to enhance their robustness.
problem Noisy neural networks struggle with random continuous noise in weights.
method Knowledge distillation combined with noise injection during training.
result Models achieve up to twice greater noise tolerance.
Langevin algorithms improve training of very deep neural networks, especially for image classification.
problem Training very deep neural networks is challenging due to increased non-linearity and the risk of getting stuck in local minima.
method Comparison of Langevin and non-Langevin algorithms for training deep neural networks, introduction of Layer Langevin algorithm.
result Langevin algorithms, especially Layer Langevin, lead to significant improvements in training deep neural networks, particularly for image classification tasks.
Recent work has established the equivalence between deep neural networks and Gaussian processes (GPs), resulting in so-called neural network Gaussian processes (NNGPs). The behaviour of these models depends on the initialisation of the corresponding network. In this work, we consider the impact of noise regularisation …
This paper examines how noise affects deep neural networks and improves their performance.
problem The impact of noise on the stability of deep ReLU neural networks for nonparametric regression.
method Investigates the optimal rate of convergence for deep ReLU neural networks under Huber loss, considering the p-th moment of noise and the smoothness of the function.
result The optimal rate of convergence cannot be achieved by ordinary least squares but can be by Huber loss with a properly chosen parameter.
Theoretical analysis of the error landscape of deep neural networks has garnered significant interest in recent years. In this work, we theoretically study the importance of noise in the trajectories of gradient descent towards optimal solutions in multi-layer neural networks. We show that adding noise (in different wa…
Improved saliency maps for deep neural networks with reduced noise.
problem Noisy explanations in Integrated Gradients for deep neural networks.
method SmoothTaylor, adaptive noising, and SmoothGrad techniques.
result SmoothTaylor and adaptive noising generate better quality saliency maps.
Image classification systems recently made a giant leap with the advancement of deep neural networks. However, these systems require an excessive amount of labeled data to be adequately trained. Gathering a correctly annotated dataset is not always feasible due to several factors, such as the expensiveness of the label…
Stochastic regularisation is an important weapon in the arsenal of a deep learning practitioner. However, despite recent theoretical advances, our understanding of how noise influences signal propagation in deep neural networks remains limited. By extending recent work based on mean field theory, we develop a new frame…
In many applications of classifier learning, training data suffers from label noise. Deep networks are learned using huge training data where the problem of noisy labels is particularly relevant. The current techniques proposed for learning deep networks under label noise focus on modifying the network architecture and…
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.
Dropout is a regularisation technique in neural network training where unit activations are randomly set to zero with a given probability \emph{independently}. In this work, we propose a generalisation of dropout and other multiplicative noise injection schemes for shallow and deep neural networks, where the random noi…
Bayesian Deep Noise Neural Network (B-DeepNoise) estimates predictive densities and uncertainty.
problem Estimating predictive densities and uncertainty in deep neural networks.
method Extends random noise to all hidden layers, using Gibbs sampling for posterior computation.
result Superior performance in prediction accuracy and uncertainty quantification.
S-SGD adds symmetrical noise to weights to avoid sharp minima in deep learning.
problem SGD does not always converge to a flat minimum, leading to poor generalization.
method Symmetrical weight noise injection in SGD.
result S-SGD outperforms conventional SGD and weight-noise injection methods in large batch training.
Deep learning approximates SPDE solutions from noise trajectories.
problem Approximating solutions to stochastic partial differential equations (SPDEs).
method Uses neural networks to approximate SPDE solutions based on noise realizations.
result Accurately estimates SPDE solutions and functionals like mean and variance.
Deep networks can overfit benignly but still be vulnerable to adversarial attacks.
problem Adversarial vulnerability of deep neural networks trained with benign overfitting.
method Investigated causes of adversarial vulnerability, identified label noise as a key factor, and explored the impact of training procedures and representation learning.
result Adversarial robustness requires more complex decision boundaries than simple ones, suggesting the need for better representation learning.
We explore the robustness of recurrent neural networks when the computations within the network are noisy. One of the motivations for looking into this problem is to reduce the high power cost of conventional computing of neural network operations through the use of analog neuromorphic circuits. Traditional GPU/CPU-cen…
This paper connects noise injection to Bayesian inference for neural networks, improving model uncertainty.
problem Improving the reliability and confidence of neural network predictions through uncertainty quantification.
method Introducing noise into neural network parameters during training and inference to estimate prediction uncertainty.
result The MCNI method outperforms baseline models in regression and classification tasks.
New method reduces uncertainty in deep neural networks with minimal computation.
problem Uncertainty in over-parameterized neural networks hinders reliability and statistical guarantees.
method Procedural-noise-correcting (PNC) predictor and resampling methods.
result Asymptotically exact-coverage confidence intervals constructed with minimal computation.
Noise balance theory explains SGD's behavior in neural networks.
problem Understanding SGD's navigation in neural network loss landscapes.
method Analyzes minibatch noise and loss function symmetries.
result Derives the stationary distribution of SGD for deep networks.
Noise Injection probes deep learning dynamics during training phases.
problem Understanding the learning mechanism of deep neural networks.
method Noise Injection Nodes (NINs) are used to perturb DNN architectures without altering the optimization algorithm.
result Distinct training phases are observed based on the scale of injected 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.
Theory explains how deep nets learn features from data.
problem Understanding how deep neural networks learn features from data.
method Developed a noise-nonlinearity phase diagram and a mechanical theory.
result Links feature learning across layers to generalization.
This research improves deep neural networks for parameter identification and prediction in stochastic Volterra integral equations.
problem Parameter identification and prediction in Volterra integral equations driven by Gaussian noise.
method Improved deep neural networks framework that incorporates inter-output relationships into the loss function.
result The framework enhances parameter estimation accuracy and provides accurate solutions for modeling stochastic systems.
Generalizes neural tangent kernel analysis for two-layer networks with noise and regularization.
problem Limitations of NTK analysis in deep learning practice.
method Generalized NTK analysis for two-layer neural networks with weight decay and gradient noise.
result Noisy gradient descent with weight decay exhibits 'kernel-like' behavior and converges linearly.
Stochastic gradient descent (SGD) forms the core optimization method for deep neural networks. While some theoretical progress has been made, it still remains unclear why SGD leads the learning dynamics in overparameterized networks to solutions that generalize well. Here we show that for overparameterized networks wit…
Colored noise improves neural network robustness against adversarial attacks.
problem Vulnerability of neural networks to adversarial perturbations.
method Injection of colored noise into network weights and activations during adversarial training.
result Our approach outperforms previous methods in terms of adversarial accuracy on CIFAR-10 and CIFAR-100 datasets.
Adaptive learning rate algorithms such as RMSProp are widely used for training deep neural networks. RMSProp offers efficient training since it uses first order gradients to approximate Hessian-based preconditioning. However, since the first order gradients include noise caused by stochastic optimization, the approxima…
Large datasets often have unreliable labels-such as those obtained from Amazon's Mechanical Turk or social media platforms-and classifiers trained on mislabeled datasets often exhibit poor performance. We present a simple, effective technique for accounting for label noise when training deep neural networks. We augment…
Recursive KalmanNet combines neural networks with Kalman filters for precise state estimation.
problem State estimation in systems with noisy measurements and non-Gaussian noise.
method Recursive KalmanNet uses a recurrent neural network to estimate states with consistent error covariance, optimizing for Gaussian negative log-likelihood.
result Recursive KalmanNet outperforms conventional Kalman filters and deep learning-based estimators in non-Gaussian noise conditions.
Paper optimizes neural network layers to reduce energy usage without sacrificing accuracy.
problem Energy consumption in deep neural networks during inference.
method Layerwise noise maximization to optimize reliability of memory elements.
result Reduces memory energy consumption by 3.3 times at equal accuracy.
The paper introduces a Hessian-based method to improve generalization in fine-tuned deep neural networks.
problem Improving generalization in fine-tuned deep neural networks, especially in noisy conditions.
method PAC-Bayesian analysis to identify a Hessian-based distance measure, proving generalization bounds, and developing an algorithm with a generalization error guarantee.
result Hessian-based distance measure correlates well with observed generalization gaps and can match the scale of these gaps in practice.
Pruning neural networks adds differential privacy noise, preserving data utility.
problem Achieving differential privacy in neural networks without sacrificing data utility.
method Proving equivalence between pruning and adding differential privacy noise to hidden-layer activations.
result Pruning can be a more effective alternative to adding differential privacy noise for neural networks.
Deep neural networks are over-parameterized, which implies that the number of parameters are much larger than the number of samples used to train the network. Even in such a regime deep architectures do not overfit. This phenomenon is an active area of research and many theories have been proposed trying to understand …
Paper presents a deep learning method for estimating asset return precision matrices in noisy financial markets.
problem Estimating precision matrices of asset returns in low signal-to-noise ratio environments.
method Non-linear factor model within deep learning framework, consistent estimator with error covariance estimator.
result Superior accuracy in simulations and empirical data.
Deep neural networks have established as a powerful tool for large scale supervised classification tasks. The state-of-the-art performances of deep neural networks are conditioned to the availability of large number of accurately labeled samples. In practice, collecting large scale accurately labeled datasets is a chal…
We present a novel methodology based on a Taylor expansion of the network output for obtaining analytical expressions for the expected value of the network weights and output under stochastic training. Using these analytical expressions the effects of the hyperparameters and the noise variance of the optimization algor…
Paper provides label complexity guarantees for deep active learning.
problem Lack of rigorous label complexity guarantees for deep active learning.
method Studied deep active learning from nonparametric classification perspective.
result Proved near-optimal label complexity guarantees for deep active learning.
Paper proposes a new framework to improve stability-based bounds in deep learning.
problem Explaining generalization in overparameterized neural networks.
method Decomposes excess risk dynamics into signal and noise components, applying stability-based bounds only to the noise.
result The decomposition framework improves stability-based bounds and explains generalization in neural networks.
Bayesian neural network models improve uncertainty quantification in multivariate regression.
problem Uncertainty quantification in multivariate regression models with heteroscedastic noise.
method Proposes Bayesian Last Layer neural network models and EM algorithms for parameter learning.
result Capable of disentangling aleatoric and epistemic uncertainty.
We examine the role of memorization in deep learning, drawing connections to capacity, generalization, and adversarial robustness. While deep networks are capable of memorizing noise data, our results suggest that they tend to prioritize learning simple patterns first. In our experiments, we expose qualitative differen…
Noise analysis detects backdoors in DNNs quickly.
problem Detecting backdoors in DNNs trained on compromised data.
method Noise-infused image titration curves to quantify robustness and detect backdoors.
result DNNs with backdoors are more sensitive to noise and reveal their targets.
Bayesian deep learning improves out-of-distribution detection but not always.
problem Improving the reliability of deep learning models in uncertain or novel data.
method Validation of likelihood-based Bayesian models for out-of-distribution detection.
result Bayesian deep learning models can marginally outperform conventional neural networks in certain conditions.
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…
We present a theoretically grounded approach to train deep neural networks, including recurrent networks, subject to class-dependent label noise. We propose two procedures for loss correction that are agnostic to both application domain and network architecture. They simply amount to at most a matrix inversion and mult…
This paper distills financial indicators into neural networks to reduce noise and improve accuracy.
problem Reduction of non-stationary noise in financial time series data.
method Co-distillation of smaller networks trained on indicators to transfer prior knowledge and reduce overfitting.
result The proposed method outperforms traditional methods in terms of speed and accuracy on real financial datasets.
Noise added to neural network features limits input reconstruction.
problem Protecting privacy during inference with deep neural networks.
method Hammersley-Chapman-Robbins bounds applied to noise added to neural network features.
result HCR bounds limit confidentiality but not accuracy.