The paper shows over-confidence in models isn't just due to over-parametrization.
problem Over-confidence in machine learning models, especially in binary classification.
method Theoretical analysis of logistic regression and other binary classification problems.
result Logistic regression is inherently over-confident in certain settings, but over-confidence is not always the case.
GUST framework improves self-training by estimating node uncertainty and generating pseudo-labels.
problem Over-confidence in pseudo-labels during self-training.
method Graph-based uncertainty-aware self-training with stochastic node labeling.
result GUST achieves state-of-the-art performance, especially in sparse labeled data settings.
The paper investigates how dataset quality and heterogeneity affect model confidence in machine learning.
problem Understanding how dataset quality and heterogeneity impact model confidence in machine learning.
method The study uses theoretical explanations and experimental demonstrations to investigate the effects of dataset size, label noise, and class heterogeneity on model confidence.
result Label noise reduces model confidence, while reduced dataset size increases it, and class heterogeneity leads to inconsistent confidence across classes.
Bayesian methods improve OoD detection in deep networks.
problem Detecting Out-of-Distribution (OoD) inputs in deep neural networks.
method Three Bayesian inference approaches applied to VAE weights.
result Improved OoD detection scores over benchmarks.
Following a long tradition of physicists who have noticed that the Ising model provides a general background to build realistic models of social interactions, we study a model of financial price dynamics resulting from the collective aggregate decisions of agents. This model incorporates imitation, the impact of extern…
Existing methods for structure discovery in time series data construct interpretable, compositional kernels for Gaussian process regression models. While the learned Gaussian process model provides posterior mean and variance estimates, typically the structure is learned via a greedy optimization procedure. This restri…
Paper introduces a method to learn physics between digital twins using imperfect models.
problem Learning physics from imperfect data and low-fidelity models.
method Bayesian Hierarchical modeling with physics-informed Gaussian processes.
result Models learning between digital twins are less uncertain than independent models but not over-confident.
We study a dynamical Ising model of agents' opinions (buy or sell) with coupling coefficients reassessed continuously in time according to how past external news (magnetic field) have explained realized market returns. By combining herding, the impact of external news and private information, we test within the same mo…
The paper proposes a method to produce well-calibrated predictions in regression tasks using maximum mean discrepancy.
problem The need for accurate uncertainty quantification in machine learning predictions.
method The method uses maximum mean discrepancy to minimize the kernel embedding measure and calibrate predictions.
result The method produces well-calibrated and sharp prediction intervals, outperforming state-of-the-art methods.
Precise estimation of uncertainty in predictions for AI systems is a critical factor in ensuring trust and safety. Deep neural networks trained with a conventional method are prone to over-confident predictions. In contrast to Bayesian neural networks that learn approximate distributions on weights to infer prediction …
Post-hoc calibration improves uncertainty under domain shift.
problem Improving uncertainty calibration under domain shift.
method Apply perturbations to validation set before post-hoc calibration.
result Perturbation step results in better calibration under domain shift.
Artificial neural networks have been successfully applied to a variety of machine learning tasks, including image recognition, semantic segmentation, and machine translation. However, few studies fully investigated ensembles of artificial neural networks. In this work, we investigated multiple widely used ensemble meth…
Class probabilities predicted by most multiclass classifiers are uncalibrated, often tending towards over-confidence. With neural networks, calibration can be improved by temperature scaling, a method to learn a single corrective multiplicative factor for inputs to the last softmax layer. On non-neural models the exist…
It has recently been shown that ReLU networks produce arbitrarily over-confident predictions far away from the training data. Thus, ReLU networks do not know when they don't know. However, this is a highly important property in safety critical applications. In the context of out-of-distribution detection (OOD) there ha…
Blade uses diffusion priors to accurately and calibratedly infer complex systems.
problem Derivative-free Bayesian inversion for high-dimensional, nonlinear problems with costly forward models.
method Blade employs an ensemble of interacting particles and diffusion models as priors, querying forward models only through evaluations.
result Blade produces well-calibrated posterior samples that existing methods cannot, improving with more iterations and particles.
Density-Softmax improves uncertainty estimation and robustness without sampling, reducing model size and latency.
problem Sampling-based uncertainty estimation methods suffer from large model size and high latency.
method Combines a Lipschitz-constrained feature extractor with the softmax layer to create a sampling-free deterministic framework.
result Density-Softmax reduces over-confidence under distribution shifts and achieves competitive results in uncertainty and robustness.
Bayesian bandits misspecification affects UX optimization, revealing new models.
problem Misspecification of value models in Bayesian bandits impacts UX optimization.
method Formulated UXO as a restless, sleeping bandit with unobserved confounders and optional stopping. Provided model extensions to address misspecifications.
result Common misspecifications lead to sub-optimal rewards, demonstrating overdispersion's effects on bandit performance.
This paper reviews metrics to assess AI model calibration accuracy.
problem AI model probabilities do not always match their true accuracy.
method Comprehensive review of 82 probability calibration metrics.
result Identified 4 classifier families and 1 object detection family of metrics.
The generalization and learning speed of a multi-class neural network can often be significantly improved by using soft targets that are a weighted average of the hard targets and the uniform distribution over labels. Smoothing the labels in this way prevents the network from becoming over-confident and label smoothing…
Characterizes uncertainty in high-dimensional linear classification models.
problem Assessing uncertainty in high-dimensional linear classification models.
method Approximate message passing algorithm for posterior marginals, closed-form formula for joint statistics.
result Closed-form formula for joint statistics between logistic classifier, Bayesian uncertainty, and ground-truth probit uncertainty.
Deep ensembles don't necessarily improve calibration in low data regimes.
problem Calibration issues in deep learning models, especially in low data regimes.
method Examination of data-augmentation, ensembling, and post-processing calibration methods.
result Standard ensembling techniques can lead to less calibrated models in low data regimes.
Bayesian neural networks improve uncertainty quantification with unlabelled data.
problem Over-confidence in predictions on covariate-shifted data.
method Approximate Bayesian inference using posterior regularisation with pseudo-labels from unlabelled data.
result Significant improvement in uncertainty quantification accuracy on covariate-shifted data.
A new AI framework reduces costs and improves performance.
problem High costs and over-confidence in AI models.
method Entropy-optimizing reformulation of Boltzmann machines.
result Cheaper, more performant AI models with validated reliability.
We propose a novel sparse spectrum approximation of Gaussian process (GP) tailored for Bayesian optimization. Whilst the current sparse spectrum methods provide desired approximations for regression problems, it is observed that this particular form of sparse approximations generates an overconfident GP, i.e. it produc…
Enhances out-of-domain calibration of neural networks.
problem Improving calibration performance of deep neural networks in out-of-domain settings.
method Consistency-guided temperature scaling (CTS) that considers style and content consistency.
result Significantly enhances out-of-domain calibration performance.
Mixup~\cite{zhang2017mixup} is a recently proposed method for training deep neural networks where additional samples are generated during training by convexly combining random pairs of images and their associated labels. While simple to implement, it has been shown to be a surprisingly effective method of data augmenta…
The study examines label smoothing to improve confidence calibration in fine-tuned LLMs.
problem Improving confidence calibration in fine-tuned large language models (LLMs) after instruction tuning.
method Examine various open-sourced LLMs, label smoothing, and custom kernel design.
result Label smoothing is effective in maintaining confidence calibration but faces challenges in large vocabulary LLMs.
CAT improves domain adaptation for fault diagnosis by calibrating teacher network predictions.
problem Performance drops in deep learning models when applied to different data distributions.
method CAT uses post-hoc calibration techniques to calibrate predictions of the teacher network during self-training.
result CAT achieves state-of-the-art performance on most transfer tasks in domain-adaptive IFD.
Study improves CNN medical image segmentation accuracy and reliability.
problem Over-confident predictions and silent failures in out-of-distribution data.
method Multi-task learning and spectral analysis of CNN feature maps.
result Joint multi-task learning models outperform dedicated models and detect OOD data more accurately.
Proposes deep quantile regression for uncertainty estimation in lesion detection.
problem Uncertainty quantification in lesion detection for critical applications.
method Quantile regression for aleatoric uncertainty, Variational AutoEncoder (VAE) with QR-VAE, binary quantile regression (BQR).
result Effective quantification of uncertainty in lesion detection and segmentation.