Study optimal ridge regularization for out-of-distribution prediction.
problem Optimal ridge regularization for predicting out-of-distribution data.
method Established conditions for optimal regularization under covariate and regression shifts, proving monotonic risk in data aspect ratio.
result Negative regularization can be optimal under shifts, even with isotropic or underparameterized training features.
PAIR optimizes machine learning models to generalize better to out-of-distribution data.
problem Optimization of machine learning models for out-of-distribution (OOD) generalization often leads to compromises that weaken robustness.
method Introduces a multi-objective optimization (MOO) perspective and a new optimization scheme called PAreto Invariant Risk Minimization (PAIR).
result PAIR improves robustness of OOD objectives by cooperatively optimizing with other objectives, yielding top OOD performances.
Paper decomposes risk into aleatoric and epistemic uncertainties and generates predictive uncertainty measures.
problem Unclear relationships between various predictive uncertainty measures in literature.
method Bayesian estimation to decompose risk into aleatoric and epistemic uncertainties, generating different predictive uncertainty measures.
result Experimental validation confirms usefulness of derived predictive uncertainty measures for detecting out-of-distribution and misclassified instances.
MaxRM uses random forests to minimize maximum risk across different environments.
problem Designing methods that generalize better to test environments with different distributions.
method Introducing variants of random forests based on the principle of MaxRM (Maximum Risk Minimization).
result Proved statistical consistency for the proposed method and provided an out-of-sample guarantee for MaxRM with regret.
Neural operators improve solving Helmholtz equation for various wave speeds.
problem Neural operators struggle with out-of-distribution scenarios for high-frequency waves.
method Proposed a subfamily of neural operators with stochastic depth for enhanced approximation of the Helmholtz equation.
result Neural operators with stochastic depth outperform standard models in out-of-distribution scenarios.
Proposes RVP to address theoretical concerns of V-REx for OOD generalization.
problem Theoretical concerns about V-REx's motivation and utility.
method Risk Variance Penalization (RVP) modifies V-REx's regularization.
result RVP discovers a robust predictor and finds invariant predictors under certain conditions.
This paper proves IRM minimizes o.o.d. risk under certain conditions.
problem Deep networks can fail to generalize to new domains with different distributions.
method Proves IRM minimizes o.o.d. risk through a bi-level optimization problem.
result IRM minimizes o.o.d. risk under specific conditions.
SLUG method detects bias and out-of-distribution content in generative models.
problem Generative models can underrepresent certain groups and fail on out-of-distribution data.
method SLUG: A new uncertainty quantification method for VAEs combining Laplace approximations and stochastic trace estimators.
result SLUG's UQ score correlates with bias and out-of-distribution content.
PEOC uses policy entropy to detect untrained states in RL.
problem Detecting untrained states in reinforcement learning for safety.
method Policy entropy based one-class classifier.
result PEOC is highly competitive and reliable.
Bayesian autoencoder detects AI safety risk of out-of-distribution inputs.
problem Detecting unreliable predictions from AI models with different distributions.
method Probabilistic, unsupervised Bayesian variational autoencoder with posterior estimation.
result Effective detection of out-of-distribution inputs in both input and latent spaces.
Work proposes a new framework to improve uncertainty estimation in deep Bayesian models.
problem Traditional training procedures underestimate uncertainty in NLMs, leading to unreliable predictions.
method Introduces a novel training framework that captures useful predictive uncertainties for out-of-distribution inputs.
result Demonstrates that traditional methods for NLMs significantly underestimate uncertainty and propose a new framework to address this issue.
MetaPhysiCa tackles robust physics-informed machine learning for OOD tasks.
problem Designing robust PIML methods for OOD forecasting tasks in physics.
method Meta-learning procedure for causal structure discovery including invariant risk minimization.
result Significantly outperforms existing PIML and deep learning methods in OOD tasks.
We introduce Invariant Risk Minimization (IRM), a learning paradigm to estimate invariant correlations across multiple training distributions. To achieve this goal, IRM learns a data representation such that the optimal classifier, on top of that data representation, matches for all training distributions. Through theo…
IRM fails to improve over standard methods in complex settings.
problem Learning invariant features for out-of-distribution generalization.
method Analysis of Invariant Risk Minimization (IRM) and related approaches under a general model.
result IRM can fail catastrophically in non-linear settings, even when test data are similar to training distribution.
New framework detects near vs. far out-of-distribution samples for AI safety.
problem Binary OOD detection fails to distinguish between semantically close and distant unknown risks.
method Ternary classification based on Low-Entropy Semantic Manifolds and Semantic Surprise Vector.
result Framework achieves state-of-the-art performance on ternary OOD detection task.
New method finds unbiased subnetworks in biased models for better OOD performance.
problem How to improve out-of-distribution generalization in deep models.
method Functional modular probing method and Modular Risk Minimization.
result Even in biased models, there are unbiased subnetworks that can achieve better OOD performance.
This work analyzes IRM and ERM from sample complexity perspective, revealing different behaviors under various distribution shifts.
problem Choosing between IRM and ERM for OOD generalization.
method Sample complexity analysis comparing IRM and ERM under different data generation mechanisms.
result IRM is preferred over ERM for certain distribution shifts, leading to better OOD generalization.
FeAT improves OOD generalization by learning richer features.
problem Improving feature learning for out-of-distribution (OOD) generalization.
method Feature Augmented Training (FeAT) iteratively augments and retains features from different subsets of training data.
result FeAT effectively learns richer features, boosting OOD performance.
Paper proposes detecting OOD examples using Gram matrices and in-distribution data.
problem Detecting OOD examples with confidence and without OOD data.
method Characterize activity patterns with Gram matrices and identify anomalies in values.
result High OOD detection rates achieved without OOD data.
WOODS benchmarks improve understanding of time series OOD generalization.
problem Limited understanding of OOD generalization in time series.
method Presented eight open-source time series benchmarks and revised OOD algorithms.
result Large room for improvement in OOD generalization algorithms for time series.
Improves regression models' performance on covariate shift.
problem Out-of-distribution generalization for regression.
method Spectrally adapting the weights of a pre-trained neural regression model.
result Spectral adaptation improves out-of-distribution performance.
Monotonic relationship found between in-distribution and out-of-distribution performance.
problem Understanding performance of machine learning models under distribution shifts.
method Analyzing ridge-regularized models and linear inverse problems under covariate shift.
result Monotonic relationship between in-distribution and out-of-distribution performance for certain models.
Paper provides a mathematical model for transformer ICL out-of-distribution generalization.
problem Understanding when transformer in-context learning can generalize beyond pre-training data.
method Minimal mathematical model of linear regression tasks with low-rank covariance matrices, analyzing distribution shifts as varying angles between subspaces.
result Transformers can generalize to all angle shifts if pre-training tasks are drawn from a union of subspaces, but not from a single Gaussian.
FOOD detects out-of-distribution samples quickly without needing OOD data.
problem Detecting out-of-distribution samples efficiently in neural networks.
method Extended DNN classifier with Gaussian layer and log likelihood ratio test.
result FOOD achieves state-of-the-art performance and is fast and applicable.
Paper tackles SCOD problem with optimal strategy and empirical validation.
problem Designing reliable prediction models abstaining from uncertain predictions.
method Bayes classifier for ID data and stochastic linear selector in 2D space.
result POSCOD method outperforms existing OOD methods.
New findings link causal models to strategic classification, improving robustness and alignment.
problem Strategic adaptation by users in classification tasks.
method Causal models to bound worst-case out-of-distribution risk.
result Causal classification optimizes classification error after adaptation under certain noise conditions.
Training on some out-of-distribution data improves generalization error before it deteriorates.
problem Generalization error improves with some out-of-distribution data but deteriorates with more.
method Synthetic datasets and deep networks on computer vision benchmarks.
result Non-monotonic trend in generalization error with OOD samples.
Ensemble models struggle with detecting mild faults.
problem Difficulty in detecting Intermediate-Severity faults due to their resemblance to normal conditions.
method Extensive experiments with ensemble models to identify and address common pitfalls.
result Designing more effective ensemble models for IS fault detection and diagnosis.
New method MRI improves machine learning models' ability to generalize to unseen data.
problem Machine learning models often fail to generalize well to out-of-distribution data.
method Introduces a new notion of invariance (MRI) and a practical version (MRI-v1) to improve model generalization.
result MRI-v1 guarantees invariant predictors and outperforms IRM-v1 in various settings.
Method enhances anomaly detection using contrastive learning and out-of-distribution data.
problem Improving anomaly detection in datasets with limited out-of-distribution data.
method Proposes a contrastive learning method that incorporates out-of-distribution data to enhance anomaly detection performance.
result The method significantly improves anomaly detection performance, even with limited out-of-distribution data.
Simple methods combine statistical tests for out-of-distribution detection.
problem Detecting data points not following the training distribution.
method Combining classical parametric tests (Rao's score test) and a typicality test.
result Combining Fisher's method of test statistics improves out-of-distribution detection accuracy.
Two-stage model improves credit scoring predictions.
problem Distribution shift in finance datasets.
method Two-stage model with out-of-distribution detection and domain knowledge.
result Highly reliable predictions for most datasets.
Bayesian Gaussian Processes layer detects out-of-distribution data in medical imaging.
problem Detecting out-of-distribution data in medical imaging tasks.
method Parameter-efficient hierarchical convolutional Gaussian Processes in Wasserstein-2 space.
result Uncertainty estimates enable superior out-of-distribution detection compared to previous methods.
Deep learning models are known to be overconfident in their predictions on out of distribution inputs. There have been several pieces of work to address this issue, including a number of approaches for building Bayesian neural networks, as well as closely related work on detection of out of distribution samples. Recent…
Modern neural networks are very powerful predictive models, but they are often incapable of recognizing when their predictions may be wrong. Closely related to this is the task of out-of-distribution detection, where a network must determine whether or not an input is outside of the set on which it is expected to safel…
Combining Bayesian deep learning and split conformal prediction affects out-of-distribution coverage.
problem Improving out-of-distribution coverage in multiclass image classification.
method Combining Bayesian deep learning with split conformal prediction methods.
result Combining methods can reduce out-of-distribution coverage in some cases.
Deep learning provides a powerful tool for machine perception when the observations resemble the training data. However, real-world robotic systems must react intelligently to their observations even in unexpected circumstances. This requires a system to reason about its own uncertainty given unfamiliar, out-of-distrib…
XEnsemble improves DNN robustness against adversarial and out-of-distribution inputs.
problem Protecting DNN models from adversarial and out-of-distribution inputs.
method Diverse input denoising verifiers and disagreement-diversity ensemble learning.
result XEnsemble achieves high defense and detection success rates.
Neural networks exhibit unimodal variance with model complexity, improving generalization.
problem The classical bias-variance trade-off does not apply to neural networks, leading to better generalization with larger models.
method Measured bias and variance of neural networks, confirmed empirically and theoretically.
result Neural networks show unimodal variance, leading to a double descent risk curve.
REx tackles distributional shift by reducing risk differences across domains.
problem Tackling distributional shift when transferring machine learning systems to real-world applications.
method Risk Extrapolation (REx) assumes training domains represent test-time variations and uses extrapolated domains to minimize risk variance.
result REx reduces sensitivity to extreme distributional shifts, including causal and anti-causal inputs.
Single deep model detects out-of-distribution data with single forward pass.
problem Detecting out-of-distribution data points in neural networks.
method Deterministic uncertainty quantification (DUQ) using gradient penalty for reliable detection.
result Single model outperforms or matches ensemble methods in out-of-distribution detection.
This paper examines how the choice of prior distribution affects likelihoods of out-of-distribution inputs in deep generative models.
problem Mismatch between prior and data distributions causes deep generative models to assign higher likelihoods to out-of-distribution inputs.
method Proposes using a mixture distribution as a prior to make likelihoods of out-of-distribution inputs more sensitive.
result A mixture prior lowers the out-of-distribution likelihood with respect to real image data sets.
Unified study of out-of-distribution generalization across 172 datasets.
problem Measuring and improving robustness of transfer learning models.
method Collect and fine-tune 31k models on 172 dataset pairs, varying architectures and settings.
result In- and out-of-distribution accuracies increase jointly but their relation is dataset-dependent.
Linear models can overfit without harming OOD generalization under certain conditions.
problem Understanding how overparameterized linear models generalize to out-of-distribution data.
method Analyzing overparameterized linear models under covariate shift, providing guarantees for OOD generalization.
result Benign overfitting occurs in standard ridge regression under OOD conditions, with specific structural conditions on target covariance.
Deep learning models are known to be overconfident in their predictions on out of distribution inputs. This is a challenge when a model is trained on a particular input dataset, but receives out of sample data when deployed in practice. Recently, there has been work on building classifiers that are robust to out of dis…
Paper proposes a framework to assess model robustness against adversarial actions.
problem Ensuring model reliability in deployment with varied adversarial conditions.
method Developed a versatile framework for evaluating SVR and relaxed optimization models' robustness.
result Demonstrates model vulnerability without requiring additional test data.
The paper shows strong correlation between in-distribution and out-of-distribution performance in various machine learning models.
problem Understanding reliability of machine learning systems in unseen environments.
method Empirical analysis of various models and distribution shifts on CIFAR-10, ImageNet, and other datasets.
result Out-of-distribution performance is strongly correlated with in-distribution performance across different models and distribution shifts.
SFP prunes ID features to improve OOD generalization without domain data.
problem Improving out-of-distribution (OOD) generalization in biased models.
method Spurious Feature-targeted model Pruning (SFP) framework.
result SFP achieves optimal OOD generalization by pruning ID features.