#distribution-shift

6 curated events

papersTODAY 04:00 UTC

Density Ratio Estimation and Importance-Weighted Regression Under Target Shift

This paper examines how to estimate density ratios and perform importance-weighted regression when the output distribution shifts between training and test data while the conditional input distribution given outputs stays the same. The authors derive optimal estimation approaches for this continuous-output target shift setting. The work falls in the statistical machine learning area of distribution shift and covariate/label shift correction.

research/machine-learning/distribution-shift/statistics

papersSEP 10 04:00 UTC

Accountable and uncertainty-aware evaluation of sensor-based AI under distribution shift

A new machine learning paper tackles the gap between training conditions and real-world deployment for sensor-based AI, where devices, personnel, and time periods all differ from the original training setup. The authors propose a staged evaluation methodology that quantifies uncertainty and captures performance degradation that conventional random train-test splits can hide. The work draws on data collected from multiple devices and subjects over nearly three years in an underground environment.

evaluation/distribution-shift/sensors/uncertainty

papersSEP 10 04:00 UTC

Researchers Propose Image Prototype Distillation for Guided Test-Time Adaptation

A new arXiv paper presents a method that distills image prototypes to guide test-time adaptation of models facing distribution shifts. The technique addresses two common failure modes in this setting: error buildup from unreliable pseudo-labels and degradation of knowledge learned during pretraining. By anchoring adaptation to distilled prototypes, the authors aim to make inference-time model updates more stable.

test-time-adaptation/distribution-shift/robustness/research

papersSEP 10 04:00 UTC

Study Compares Retraining Policies for Subgroup Disparity Under Data Drift

A new arXiv paper examines how the choice of retraining policy affects subgroup error rates in deployed classifiers as data distributions drift. The authors run paired comparisons of complete scheduled retraining against loss-triggered and subgroup-gap-triggered approaches, tracking cumulative subgroup disparity across model sequences, including gaps between updates. The work frames retraining timing as a question of fairness measurement rather than accuracy alone.

fairness/retraining/distribution-shift/evaluation

papersSEP 11 04:00 UTC

New Framework Quantifies Covariate and Concept Shifts in ML Generalization

A new arXiv paper proposes a general approach to measuring how covariate and concept shifts affect machine learning generalization. The authors argue that existing learning bound theory covers only narrow, idealized settings and cannot be estimated from data. Their framework aims to make distribution shift analysis broadly applicable and computable from samples.

research/distribution-shift/generalization/theory

papersSEP 12 04:00 UTC

arXiv paper proposes importance weighting for unlabeled-unlabeled learning under distribution shift

A new arXiv preprint addresses unlabeled-unlabeled (UU) learning, a setting where a binary classifier is trained from two unlabeled datasets that have differing class priors. The authors introduce importance weighting to handle distribution shift in this framework, which generalizes approaches such as positive-unlabeled learning. The work is listed as a cross-submission in the cs.AI category.

research/weak-supervision/distribution-shift/machine-learning