Shifts dataset evaluates uncertainty in real-world tasks across modalities.
problem Lack of standard datasets for evaluating uncertainty estimation and robustness to distributional shift.
method Proposes Shifts Dataset for evaluation of uncertainty estimates and robustness to distributional shift across tabular, audio, text, and sensor data.
result Baseline results for tabular weather prediction, machine translation, and SDC vehicle motion prediction.
Paper proposes a framework to detect distribution shifts using embedding space geometry.
problem Detecting distribution shifts in candidate datasets to improve model generalizability.
method Non-parametric framework using embedding space geometry for two tests: robustness boundary and in-distribution/out-of-distribution classification.
result Both tests successfully detect distribution shifts in various scenarios for both synthetic and real-world datasets.
New framework identifies worst-case shifts for predictive resource allocation models.
problem Identifying harmful shifts in predictive models for resource allocation.
method Hierarchical model structure and submodular optimization for worst-case loss.
result Empirical evidence shows divergent worst-case shifts identified by different metrics.
Extends Shifts dataset for MS lesion segmentation and marine vessel power estimation.
problem Distributional shift in training and deployment data for ML models.
method Develops new datasets for high-risk industrial applications.
result Demonstrates robustness and uncertainty estimation in new industrial tasks.
Covariate shift relaxes the widely-employed independent and identically distributed (IID) assumption by allowing different training and testing input distributions. Unfortunately, common methods for addressing covariate shift by trying to remove the bias between training and testing distributions using importance weigh…
Task shift from classification to regression is possible in overparameterized linear models with limited additional data.
problem Transferability of latent knowledge from classification to regression in overparameterized linear models.
method Investigation of task shift in overparameterized linear regression, zero-shot and few-shot cases, with a focus on minimum-norm interpolation.
result Minimum-norm interpolators can transfer latent knowledge from classification to regression with limited additional data.
Public pretraining improves private model training even in extreme distribution shift scenarios.
problem Improving private model training accuracy in settings with large distribution shift.
method Empirical evaluation and theoretical explanation of public representations improving private training accuracy.
result Public representations can improve private training accuracy by up to 67% over private training from scratch in settings with large distribution shift.
We created financial benchmarks for distribution shifts in crude oil prices and volatility.
problem Scarcity of task-labeled time-series benchmarks in finance.
method Transformed asset price data into volatility proxies, generated task labels based on distribution shifts, and made datasets publicly available.
result Inclusion of task labels improves continual learning algorithms' performance on real-world data.
Improves model calibration and selection in unsupervised domain adaptation.
problem Distribution shifts in unsupervised domain adaptation.
method Developed a novel importance weighted group accuracy estimator.
result Improves state-of-the-art performances by 22% in model calibration and 14% in model selection.
New bounds for contrastive learning handle domain shifts and generalization.
problem Domain shifts and generalization challenges in downstream tasks.
method Novel generalization bounds accounting for both domain shift and generalization.
result Performance of contrastively learned representations depends on statistical discrepancy between pretraining and downstream distributions.
RAVEN improves weak-to-strong generalization under distribution shifts.
problem Weak models fail to supervise strong models effectively under distribution shifts.
method RAVEN dynamically learns optimal combinations of weak models and strong model parameters.
result RAVEN outperforms existing methods by over 30% on out-of-distribution tasks.
Conformal prediction fails under severe feature turnover in COVID-19 supply chain tasks.
problem Dealing with distribution shift in conformal prediction models.
method Using COVID-19 as a natural experiment across 8 supply chain tasks, analyzing SHAP explanations.
result Coverage drops vary widely (0% to 86.7%) and correlate with single-feature dependence.
MAGDiff detects data shifts in neural networks without retraining.
problem Neural networks' sensitivity to data distribution shifts.
method Extracts MAGDiff representations from neural networks to detect shifts.
result MAGDiff representations improve data set shift detection.
Proposes a FoE prior for improving CNN performance in distribution shifts.
problem Improving CNN performance in image analysis tasks with distribution shifts.
method Uses a field-of-experts (FoE) prior to match feature distributions of test and training images.
result Outperforms previous TTA methods in lesion segmentation and most healthy tissue segmentation tasks.
Improves overparameterized models' robustness to distribution shifts.
problem Accuracy drop on testing distributions different from training.
method Importance tempering to improve decision boundaries.
result State-of-the-art results on worst group classification tasks.
This work evaluates graph models' robustness to structural distributional shifts.
problem Evaluating graph models' robustness to structural distributional shifts.
method Proposes a general approach for inducing diverse distributional shifts based on graph structure.
result Simple models often outperform more sophisticated methods on structural distributional shifts.
Proposes a method to improve regression model performance with limited target data using fused-regularizer.
problem Model shifts and covariate shifts in high-dimensional regression.
method Two-step method with fused-regularizer to leverage source data for target task.
result Robust to covariate shifts, minimax-optimal under certain conditions, and validated by numerical tests.
Novel hyperparameter optimization for target tasks under covariate shift.
problem Hyperparameter optimization under multi-source covariate shift.
method Construct variance reduced estimator to unbiasedly approximate target objective; propose no-regret hyperparameter optimization procedure.
result Proposed framework broadens applications of automated hyperparameter optimization.
New framework identifies and reduces errors in machine learning under distribution shift.
problem Errors in machine learning models when distributions change.
method Developed a principled framework to characterize and eliminate epistemic errors in imperfect multitask learning.
result Provided a decompositional epistemic error bound for general settings of distribution shift.
FDN improves probabilistic regressors' adaptability to distribution shifts.
problem Overconfidence in modern probabilistic regressors under distribution shift.
method FDN uses input-conditioned distributions over network weights, trained with a Monte Carlo beta-ELBO objective.
result FDN produces predictive mixtures whose dispersion adapts to the input, providing shift-aware uncertainty.
Simple method improves uncertainty estimation for distribution shifts.
problem Improving uncertainty estimation in deep image classification under distribution shifts.
method Exposing original model to corrupted images and performing simple statistical calibration.
result Superior performance on various distribution shifts and unsupervised domain adaptation tasks.
The paper analyzes how combining samples from two tasks can improve performance, especially in high dimensions.
problem Understanding when combining samples from two related tasks outperforms learning with one task alone.
method Applying random matrix theory to high-dimensional linear regression, focusing on proportional sample size increases.
result Precise high-dimensional asymptotics for bias and variance of HPS estimator, showing phase transitions in transfer performance.
Improved vehicle motion prediction with uncertainty estimation.
problem Robust motion prediction for autonomous vehicles, especially under distributional shift.
method Presented an approach significantly improving the benchmark and taking 2nd place on the leaderboard.
result Significantly improved motion prediction and uncertainty measurement.
DORO improves DRO's performance and stability in tasks with subpopulation shift.
problem DRO's poor performance and instability in tasks with subpopulation shift.
method DORO, a refined risk function that prevents overfitting to outliers.
result DORO improves DRO's performance and stability on large modern datasets.
Paper quantifies label shift with robustness guarantees using distribution feature matching.
problem Estimating target label distribution under label shift.
method Distribution feature matching (DFM) framework and robustness analysis.
result General performance bound and robustness analysis in misspecified settings.
We consider the problem of function estimation in the case where the data distribution may shift between training and test time, and additional information about it may be available at test time. This relates to popular scenarios such as covariate shift, concept drift, transfer learning and semi-supervised learning. Th…
Reweighting training data to better represent new tasks.
problem Deploying machine learning models to new tasks is challenging due to training data distribution.
method Formulate an exponential tilt distribution shift model and learn train data importance weights to minimize KL divergence.
result The learned train data weights improve target performance evaluation, fine-tuning, and model selection.
New method adapts to structural shifts in graph data for better label prevalence estimation.
problem Structural shifts in graph data affect label prevalence estimation.
method Importance sampling variant of KDEy quantification approach.
result Adapts to structural shifts and outperforms standard approaches.
Transformers show better in-context learning resilience under distribution shifts than simple MLPs.
problem Understanding in-context learning under varying distribution shifts.
method Comparing transformers and set-based MLPs on linear regression tasks.
result Transformers better emulate OLS performance and exhibit better resilience to mild distribution shifts.
Study improves conformal prediction for EEG classification in healthcare, enhancing coverage.
problem Uncertainty quantification in clinical predictions, especially in distribution-shifted settings.
method Personalized calibration strategies to improve coverage of prediction sets.
result Coverage improved by over 20 percentage points with comparable prediction set sizes.
Object-centric learning improves generalization and robustness in multi-object scenes.
problem Improving generalization and robustness in neural networks for scenes with multiple objects.
method Training state-of-the-art unsupervised models on multi-object datasets and evaluating segmentation metrics and downstream tasks.
result Object-centric representations are useful for downstream tasks and generally robust to most distribution shifts affecting objects, but less so for less structured shifts.
New approach combines invariance and information bottleneck for OOD generalization.
problem OOD generalization failures in classification tasks.
method Revisit linear regression tasks, prove information bottleneck constraint necessary, propose combined approach.
result Combined invariance and information bottleneck approach improves OOD generalization.
Transformers learn to adapt to different task difficulties and resist distribution shifts.
problem Understanding and optimizing a Transformer's performance across various task difficulties and distribution shifts.
method Analyzing a pretrained Transformer on a mixture distribution of tasks, proving optimal convergence rates.
result Transformers achieve optimal convergence rates on tasks of specific difficulty levels, robust to distribution shifts.
DSLOB creates synthetic LOB data for benchmarking forecasting algorithms under distributional shifts.
problem Challenges in dealing with out-of-distribution limit order book data.
method Multi-agent market simulator to create labeled synthetic LOB dataset with and without market stress.
result Demonstrates the need for robust forecasting algorithms to handle distributional shifts.
JAWS audits predictive uncertainty under covariate shift using jackknife+ weighted methods.
problem Auditing predictive uncertainty under data distribution shifts.
method JAW and JAWA methods for distribution-free uncertainty quantification.
result JAW relaxes the jackknife+'s assumption of data exchangeability for covariate shift.
MADOD meta-learns invariant features for OOD detection across unseen domains.
problem Simultaneous covariate and semantic shifts in real-world machine learning applications.
method Meta-learning and G-invariance to learn robust, domain-invariant features.
result Superior performance in semantic OOD detection across unseen domains.
DoubleAdapt improves stock trend forecasting by adapting models to evolving data.
problem Incremental learning for stock trend forecasting is challenging due to distribution shifts.
method DoubleAdapt framework with two adapters for data and model adaptation.
result DoubleAdapt achieves state-of-the-art predictive performance on real-world stock datasets.
Offline RL with pre-trained features amplifies errors even under mild shifts.
problem Sample-efficient offline RL with pre-trained features under mild distribution shift.
method Empirical study of offline RL with pre-trained neural representations.
result Substantial error amplification occurs even with pre-trained features, requiring stronger conditions for successful offline RL.
CP improves robustness against distribution shift using physics-informed structural causal models.
problem Uncertainty in machine learning predictions under distributional shift.
method Physics-informed structural causal model (PI-SCM) to upper bound coverage difference.
result PI-SCM improves coverage robustness across confidence levels and test domains.
CoDrug uses KDE to create valid prediction sets for drug molecules under covariate shift.
problem Creating reliable uncertainty estimates for drug properties from computational models.
method CoDrug employs an energy-based model and KDE to assess and rectify distribution shift.
result CoDrug reduces the coverage gap by over 35% compared to non-adjusted conformal prediction sets.
Boosted Control Functions improve prediction under distributional shifts.
problem Prediction under distributional shifts in the presence of hidden confounding.
method Boosted Control Function (BCF) and ControlTwicing algorithm.
result BCF allows for distribution generalization and invariance under nonlinear, non-identifiable structural functions.
Study on neural scaling laws for solving linear systems in-context.
problem Theoretical guarantees for solving linear systems using a linear transformer architecture.
method Neural scaling laws and task diversity for in-domain and out-of-domain generalization.
result Novel notion of task diversity for necessary and sufficient condition of generalization under task shifts.
CTRF combines logged data and randomized experiments for robust prediction.
problem Robust prediction models to handle distributional shifts between training and testing data.
method CTRF uses existing training data and a small amount of randomized experiment data to train a robust model.
result CTRF produces robust predictions and outperforms baseline methods in the presence of feature shifts.
Proposes a framework to assess model robustness to dataset shifts.
problem Evaluating model robustness to changes in setting or population.
method Derives a debiased estimator for analyzing performance on worst-case distributions.
result Demonstrates the estimator can account for realistic shifts in complex distributions.
Adaptive monitoring for AI systems detects and diagnoses shifts in data distribution.
problem Continuous monitoring of AI systems to detect and address unsafe behavior.
method Weighted-conformal martingales (WCTMs) for online monitoring of AI systems.
result Improved performance over state-of-the-art baselines on real-world datasets.
Transformers learn linear models in-context without updates.
problem Understanding how transformers mimic linear models in-context.
method Gradient flow on linear regression tasks with random initialization.
result Transformers achieve prediction error competitive with best linear predictors.
Imitation learning has gained immense popularity because of its high sample-efficiency. However, in real-world scenarios, where the trajectory distribution of most of the tasks dynamically shifts, model fitting on continuously aggregated data alone would be futile. In some cases, the distribution shifts, so much, that …
New methods for Bayesian inference using mean shift particle systems.
problem Approximating expectations with unnormalized densities in Bayesian inference.
method Mean shift interacting particle systems that minimize maximum mean discrepancy (MMD).
result Mean shift interacting particle systems converge quickly and capture complex distributions.