DBPA assesses LLM perturbations using frequentist hypothesis testing.
problem Quantifying input perturbation impacts on LLM outputs.
method DBPA reformulates perturbation analysis as frequentist hypothesis testing, using Monte Carlo sampling for empirical null and alternative distributions.
result DBPA provides interpretable p-values and scalar effect sizes for LLM perturbations.
Learnable token perturbations boost extrapolation in LLMs.
problem Limited flexibility of current discrete perturbations in large language models.
method Learnable continuous latent vector transformations in embedding space, unbiased estimating equations, stochastic gradient descent optimization.
result Significant gains in out-of-domain settings over state-of-the-art methods.
New method defends deep nets against large perturbations perceptible to humans.
problem Vulnerability of deep nets to adversarial attacks with perceptible but not changing predictions.
method Oracle-Aligned Adversarial Training (OA-AT) to align network predictions with Oracle's.
result Achieves state-of-the-art performance at large perturbation bounds (L-inf of 16/255 on CIFAR-10).
This work analyzes the stability of graph filters under large perturbations.
problem Stability of graph filters under large edge rewires.
method Proves a bound on stability using frequency response and community structure.
result Graph filter stability depends on perturbation to community structure.
New method makes deep nets robust to large perturbations.
problem Making deep nets robust to large perturbations.
method Simple weight initialization and rigorous evaluation on MNIST and CIFAR10.
result Learning on higher levels of noise is feasible with new initialization.
Adversarial fog tests autonomous navigation models.
problem Neural networks are fooled by adversarial perturbations, but fog naturally creates similar perturbations.
method Introduced a new type of adversarial perturbation using generative models and Cycle-Consistent Generative Adversarial Networks.
result Generated adversarial fog images help test autonomous navigation models.
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.
Deep learning detects cloud changes due to human aerosols.
problem Uncertainty in the effect of anthropogenic aerosols on cloud properties and Earth's energy balance.
method Deep convolutional neural networks to analyze cloud images.
result Identified and characterized specific cloud perturbations due to human aerosols.
A fast method computes class-specific adversarial perturbations for deep networks.
problem Computing robust adversarial perturbations for deep networks.
method Linear function of weights, no training data, no hyper-parameters.
result Obtains 34% to 51% fooling rate on ImageNet, transfers across models.
We show the existence of isoperimetric regions of sufficiently large volumes in general asymptotically hyperbolic three manifolds. Furthermore, we show that large coordinate spheres in compact perturbations of Schwarzschild-anti-deSitter are uniquely isoperimetric. This is relevant in the context of the asymptotically …
Lectures on deep learning properties in infinite and large-width networks.
problem Understanding deep neural networks in extreme width conditions.
method Analysis of random deep neural networks, connections to linear models, kernels, and Gaussian processes, perturbative and non-perturbative treatments.
result Properties and behaviors of deep neural networks in the infinite-width limit and large-width regime.
Improved algorithm speeds up generation of universal adversarial perturbations.
problem Slow generation of universal adversarial perturbations.
method Optimized algorithm based on orientation of perturbation vectors.
result Significantly faster generation of universal perturbations with higher fooling rates.
Tseytlin has recently proposed that an action functional exists whose gradient generates to all orders in perturbation theory the Renormalization Group (RG) flow of the target space metric in the worldsheet sigma model. The gradient is defined with respect to a metric on the space of coupling constants which is explici…
Paper develops a new kernel approximation framework.
problem High time and space complexity of kernel methods for large datasets.
method Perturbation-based kernel approximation framework using classical perturbation theory.
result Framework generalizes and improves upon existing methods.
Stochastic neural net weights are used in a variety of contexts, including regularization, Bayesian neural nets, exploration in reinforcement learning, and evolution strategies. Unfortunately, due to the large number of weights, all the examples in a mini-batch typically share the same weight perturbation, thereby limi…
Perturbation theory improves nonparametric instrumental variable estimation accuracy.
problem Improving nonparametric instrumental variable estimation accuracy in high-dimensional settings.
method Perturbative approach based on physics perturbation theory, extending kernel ridge methods with higher-order corrections.
result First-order perturbative corrections reduce prediction error by up to 99% in high-dimensional ill-defined cases.
Thanks to recent advances in deep neural networks (DNNs), face recognition systems have become highly accurate in classifying a large number of face images. However, recent studies have found that DNNs could be vulnerable to adversarial examples, raising concerns about the robustness of such systems. Adversarial exampl…
This paper proposes a new algorithm for controlling classification results by generating a small additive perturbation without changing the classifier network. Our work is inspired by existing works generating adversarial perturbation that worsens classification performance. In contrast to the existing methods, our wor…
We consider uncorrelated Stein-Stein, Heston, and Hull-White models and their perturbations by compound Poisson processes with jump amplitudes distributed according to a double exponential law. Similar perturbations of the Black-Scholes model were studied by S. Kou. For perturbed stochastic volatility models, we obtain…
ContrastiveVI+ models CRISPR screens with noisy guide efficiency.
problem Noisy guide efficiency in CRISPR screens.
method Generative modeling framework that disentangles perturbation-induced from shared variations.
result ContrastiveVI+ better recovers perturbation-induced variations and identifies cells without edits.
EdgeFool generates adversarial images to mislead classifiers.
problem Misleading classifiers with adversarial images.
method Trains a fully convolutional neural network to generate perturbations that enhance image details and mislead classifiers.
result EdgeFool outperforms other adversarial methods on various classifiers and datasets.
SmoothLLM defends LLMs from jailbreaking attacks by randomly perturbing inputs.
problem Adversaries can fool large language models into generating objectionable content.
method SmoothLLM randomly perturbs multiple copies of a prompt and aggregates predictions to detect adversarial inputs.
result SmoothLLM sets the state-of-the-art for robustness against various jailbreak attacks.
As increasing amounts of sensitive personal information is aggregated into data repositories, it has become important to develop mechanisms for processing the data without revealing information about individual data instances. The differential privacy model provides a framework for the development and theoretical analy…
Gradient noise improves privacy-protected optimization performance.
problem Improving privacy in convex optimization while maintaining utility.
method We analyze the effect of gradient perturbation on differentially private convex optimization, focusing on expected curvature.
result Gradient perturbation can achieve a significantly improved utility guarantee for differentially private convex optimization.
New geometric interpretation explains over-parameterized models and adversarial perturbations.
problem Geometric understanding of over-parameterized regression and adversarial perturbations.
method Alternative geometric interpretation of regression in feature space.
result Adversarial perturbations are a natural feature of biased models due to underlying geometry.
TULiP estimates uncertainty for deep learning models safely.
problem Reliable uncertainty estimation for deep learning models in the open world.
method TULiP considers a hypothetical perturbation, bounds its effect, and computes uncertainty from sampled predictions.
result TULiP achieves state-of-the-art performance in OOD detection benchmarks.
Semantify-NN verifies neural network robustness against semantic perturbations.
problem Verifying robustness of neural networks against semantic adversarial attacks.
method Inserting semantic perturbation layers (SP-layers) into neural networks to verify robustness.
result Semantify-NN significantly improves robustness verification performance over ℓp-norm-based methods. Noise injection before gradient steps helps in regularization for neural networks.
problem Improving generalization in overparametrized neural networks.
method Injecting small noise perturbations before computing gradient steps, especially in layer-wise fashion.
result Small noise perturbations can explicitly regularize neural networks without variance explosion.
We consider a classifier whose test set is exposed to various perturbations that are not present in the training set. These test samples still contain enough features to map them to the same class as their unperturbed counterpart. Current architectures exhibit rapid degradation of accuracy when trained on standard data…
New method proves instability of naked singularity and censors it.
problem Proving instability and censoring naked singularity.
method Einstein-scalar field system, hyperbolic short-pulse method, non-perturbative elliptic arguments.
result Tiny anisotropic perturbation leads to anisotropic apparent horizon censoring the naked singularity.
Within the unmanageably large class of nonconvex optimization, we consider the rich subclass of nonsmooth problems that have composite objectives---this already includes the extensively studied convex, composite objective problems as a special case. For this subclass, we introduce a powerful, new framework that permits…
Improved neural network robustness with instance-specific perturbation margins.
problem Adversarial training fails to generalize well to unperturbed test set.
method Instance adaptive adversarial training with sample-specific perturbation margins.
result Test accuracy improves with a marginal drop in robustness.
High sensitivity of neural networks against malicious perturbations on inputs causes security concerns. To take a steady step towards robust classifiers, we aim to create neural network models provably defended from perturbations. Prior certification work requires strong assumptions on network structures and massive co…
We continue the study of statistical/computational tradeoffs in learning robust classifiers, following the recent work of Bubeck, Lee, Price and Razenshteyn who showed examples of classification tasks where (a) an efficient robust classifier exists, in the small-perturbation regime; (b) a non-robust classifier can be l…
A new method improves language model extrapolation without changing training.
problem Improving language model performance on sequences not seen during training.
method Perturbation-based training procedure for language models.
result The method consistently improves out-of-support prediction.
New metrics improve scRNA-seq perturbation modeling by reducing mode collapse.
problem Outperformed by simple mean prediction in scRNA-seq perturbation modeling.
method Introduce DEG-aware metrics (WMSE, Rw2(Δ)) and negative/positive baselines. result WMSE loss function reduces mode collapse and improves model performance.
Adversarial training leads to large generalization gap, decomposed into bias and variance.
problem Understanding the large generalization gap in adversarially trained models.
method Bias-Variance decomposition of test risk as a function of adversarial perturbation radius.
result Bias increases monotonically with adversarial perturbation radius and is dominant in test risk.
PRoA assesses deep learning robustness against practical functional perturbations.
problem Inadequate practical robustness verification methods for deep learning systems.
method Probabilistic robustness assessment based on adaptive concentration.
result Statistical guarantees on probabilistic robustness against functional perturbations.
In this paper, we propose an efficient Monte Carlo implementation of non-linear FBSDEs as a system of interacting particles inspired by the ideas of branching diffusion method. It will be particularly useful to investigate large and complex systems, and hence it is a good complement of our previous work presenting an a…
Develops fair classifiers robust to training distribution perturbations.
problem Ensuring fairness in classifiers robust to training data perturbations.
method Formulates a min-max objective function to minimize distributionally robust training loss while maintaining fairness for perturbed distributions. Uses an iterative online learning algorithm to find a fair and robust classifier.
result Our classifier maintains fairness and accuracy for a wide range of perturbations compared to state-of-the-art fair classifiers.
Constructing an efficient parameterization of a large, noisy data set of points lying close to a smooth manifold in high dimension remains a fundamental problem. One approach consists in recovering a local parameterization using the local tangent plane. Principal component analysis (PCA) is often the tool of choice, as…
A machine learning model that generalizes well should obtain low errors on unseen test examples. Thus, if we know how to optimally perturb training examples to account for test examples, we may achieve better generalization performance. However, obtaining such perturbation is not possible in standard machine learning f…
Paper addresses eigenvector perturbation in small eigen-gap scenarios.
problem Fine-grained behavior of eigenvectors in the presence of small eigen-gaps.
method Develops de-biased estimators for linear functions of an unknown eigenvector.
result Achieves minimax lower bounds for a family of scenarios, even with small eigen-gaps.
Deep Convolutional Networks (DCNs) have been shown to be sensitive to Universal Adversarial Perturbations (UAPs): input-agnostic perturbations that fool a model on large portions of a dataset. These UAPs exhibit interesting visual patterns, but this phenomena is, as yet, poorly understood. Our work shows that visually …
New attack tricks certifiably robust models into mislabeling images.
problem Defeating certified defenses against adversarial examples.
method Spoofed robustness certificates and large perturbations.
result Certifiably robust models can be fooled by large perturbations.
We show that under suitable non-degeneracy conditions, complete gradient flow lines of the scalar curvature functional of a riemannian manifold perturb into eternal forced mean curvature flows with large forcing term.
Recent research has demonstrated the brittleness of machine learning systems to adversarial perturbations. However, the studies have been mostly limited to perturbations on images and more generally, classification that does not deal with temporally varying inputs. In this paper we ask "Are adversarial perturbations po…
State-of-the-art machine learning models frequently misclassify inputs that have been perturbed in an adversarial manner. Adversarial perturbations generated for a given input and a specific classifier often seem to be effective on other inputs and even different classifiers. In other words, adversarial perturbations s…