This work proposes using Conformal Prediction to improve OOD detection scores and vice versa.
problem Improper evaluation of OOD detection scores due to finite sample size.
method Defining new conformal AUROC and FRP@TPR95 metrics and using OOD scores as non-conformity scores.
result Improved evaluation metrics and better interpretation of OOD scores.
INK scores improve OOD detection for classifiers.
problem Detecting out-of-distribution inputs for classification models.
method INK scores operate on constrained latent embeddings modeled as a mixture of hyperspherical embeddings, optimizing in modern neural networks.
result INK establishes a new state-of-the-art in OOD detection.
A new OOD detector using an overlap index improves accuracy without high computational costs.
problem Effective OOD detection for machine learning models in open-world scenarios.
method Proposes an overlap index-based confidence score function for OOD detection.
result The proposed method achieves competitive accuracy with lower computational costs compared to state-of-the-art detectors.
Paper proposes contrastive training to improve OOD detection without needing explicit OOD examples.
problem Improving reliable detection of out-of-distribution inputs for machine learning systems.
method Contrastive training approach that doesn't require explicit OOD examples, using CLP score.
result Contrastive training significantly improves OOD detection performance on benchmarks, especially in near OOD classes.
P-OCS detects OOD samples in a low-dimensional subspace, outperforming existing methods.
problem Efficient OOD detection for deep learning models in open-world environments.
method P-OCS operates in the orthogonal complement of the principal subspace, applying a single projected perturbation.
result P-OCS achieves state-of-the-art OOD detection with negligible computational cost and without requiring model retraining.
Framework uses human feedback to safely set OOD detection thresholds, reducing false positives.
problem Challenges in setting OOD detection thresholds for safety-critical applications.
method Mathematically grounded framework leveraging expert feedback to dynamically update thresholds.
result Guaranteed to meet FPR constraint while minimizing human feedback, maintaining FPR at most 5%.
BoC probe assesses neural network confidence coherence, revealing architecture-specific uncertainty.
problem Poor calibration and OOD detection in neural networks.
method Bag-of-Coins (BoC) probe compares softmax confidence to pairwise dominance probabilities.
result BoC reveals clear ID/OOD separation for some architectures but not others.
Proposes Likelihood Regret for VAEs to improve OOD detection.
problem VAEs can assign high likelihoods to OOD samples, making traditional likelihood thresholds unreliable.
method Introduces Likelihood Regret, a new OOD score for VAEs.
result Empirical results show Likelihood Regret outperforms existing methods for VAEs.
Gaussian process detects OOD data without needing OOD samples.
problem Overconfident predictions from DNNs on OOD data.
method Gaussian process for uncertainty quantification and decision boundary.
result Outperforms state-of-the-art methods in OOD detection.
Bayesian Neural Networks improve OOD detection with limited data.
problem Limited training data hinders reliable OOD detection.
method Bayesian Neural Networks with expected logit vectors.
result Bayesian methods outperform deterministic methods in small data settings.
A new OOD detection method OTOD uses optimal transport theory to improve model performance.
problem Detecting unknown samples in real-world machine learning models.
method OTOD uses optimal transport theory to calculate an OOD score combining features, logits, and softmax probability space.
result OTOD outperforms state-of-the-art methods by significant margins on benchmarks.
Geometrically, high-likelihood regions in DGMs are unlikely to generate OOD data.
problem The paradox of high-likelihood OOD detection in deep generative models.
method Local intrinsic dimension estimation to identify high-likelihood regions that do not generate OOD data.
result A method pairing likelihoods and LID estimates for reliable OOD detection.
The study proposes a framework to accept OOD data based on competence scores.
problem Silent failures in Domain Generalization where models reject OOD data without proper justification.
method A learning to reject framework using proxy incompetence scores to predict trustworthiness.
result Increasing incompetence scores are predictive of reduced accuracy, but not always favorable for accuracy/rejection trade-off.
ProSub uses angles in feature space to classify data as in- or out-of-distribution.
problem Open-set semi-supervised learning with unknown classes.
method Probabilistic approach based on angles in feature space, estimating conditional distributions of scores.
result ProSub achieves state-of-the-art performance on benchmark problems.
Proposes a new method to measure epistemic uncertainty in Bayesian neural networks.
problem Measuring epistemic uncertainty in Bayesian neural networks for out-of-distribution detection.
method Proposes measuring disagreement between logits and their pre-softmax counterparts as an epistemic uncertainty measure.
result Proposed epistemic uncertainty scores outperform mutual information and equal predictive entropy performance.
Universal approach combines OOD detection scores for robustness.
problem Combining diverse OOD detection scores for robustness.
method Quantile normalization to p-values, meta-analysis, probabilistic interpretation.
result Significantly improved robustness and performance across diverse OOD detection scenarios.
WOOD detects out-of-distribution samples using Wasserstein distance.
problem Detecting samples from different distributions in neural networks.
method WOOD defines a Wasserstein-distance-based score to evaluate dissimilarity and solves an optimization problem.
result WOOD consistently outperforms other OOD detection methods.
Framework detects out-of-distribution inputs in regression and survival analysis.
problem Limited OOD detection for regression and survival analysis.
method Model-aware and subspace-aware variable prioritization.
result Consistent improvements over existing methods in synthetic and real data.
Igeood detects out-of-distribution samples using information geometry.
problem Out-of-distribution (OOD) detection in machine learning systems.
method Igeood uses the Fisher-Rao geodesic distance to detect OOD samples from any pre-trained neural network.
result Igeood outperforms state-of-the-art methods on various network architectures and datasets.
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.
New method improves OOD detection by integrating diffusion models into discriminator models.
problem Overconfidence in discriminator models leads to poor OOD detection.
method Integrates diffusion models into discriminator and generation models to mitigate overconfidence.
result Demonstrates significant improvement in AUROC scores for challenging datasets.
TopoGeoScore selects robust checkpoints using only source-domain representations.
problem Selecting robust checkpoints without target-domain labels or samples.
method Constructs class-conditional mutual k-nearest-neighbour graphs and extracts three interpretable signals.
result Source representations contain measurable global-local-topological evidence of robustness.
Likelihood-based generative models are a promising resource to detect out-of-distribution (OOD) inputs which could compromise the robustness or reliability of a machine learning system. However, likelihoods derived from such models have been shown to be problematic for detecting certain types of inputs that significant…
This paper combines existing OOD detection methods to improve overall performance.
problem Improving robustness of neural networks in safety-critical applications.
method Integrates four strategies for combining multiple OOD detection scores.
result Enhanced OOD detection through multi-dimensional evaluation metrics.
As deep learning methods form a critical part in commercially important applications such as autonomous driving and medical diagnostics, it is important to reliably detect out-of-distribution (OOD) inputs while employing these algorithms. In this work, we propose an OOD detection algorithm which comprises of an ensembl…
Improved OOD detection across various shifts using multi-encoder fusion of RDMs.
problem Out-of-distribution detection across multiple types of distribution shifts.
method Statistical identification of encoder sensitivity, EncMin2L fusion, and Tippett minimum combination.
result Achieves AUROC ≥ 0.94 across four shift types, outperforming state-of-the-art detectors.
OMASGAN generates anomalous samples on distribution boundary to improve anomaly detection.
problem Missed anomalies and low AD performance due to OoD samples in generative models.
method OMASGAN generates anomalous samples on the estimated distribution boundary using a GAN approach, refining AD models.
result OMASGAN improves AD performance by at least 0.24 and 0.07 points on average on MNIST and CIFAR-10 datasets.
Out-of-distribution (OOD) detection approaches usually present special requirements (e.g., hyperparameter validation, collection of outlier data) and produce side effects (e.g., classification accuracy drop, slower energy-inefficient inferences). We argue that these issues are a consequence of the SoftMax loss anisotro…
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.
Bayesian nonparametric models improve OOD detection, especially with complex covariance structures.
problem Improving out-of-distribution detection methods, especially in complex scenarios.
method Proposes Bayesian nonparametric mixture models with hierarchical priors that generalize the Mahalanobis distance score.
result Bayesian nonparametric methods outperform existing OOD methods, especially in complex scenarios.
Discriminative neural networks offer little or no performance guarantees when deployed on data not generated by the same process as the training distribution. On such out-of-distribution (OOD) inputs, the prediction may not only be erroneous, but confidently so, limiting the safe deployment of classifiers in real-world…
This paper examines how optimization methods affect the reliability of detecting inputs outside a model's training distribution.
problem The unreliability of deep neural networks on out-of-distribution inputs.
method Analysis of optimization methods' impact on OOD detection approaches.
result Optimization methods significantly influence the robustness of OOD detection approaches.
Study complex-valued VAEs for radar OOD detection.
problem Detecting out-of-distribution signals in complex radar environments.
method Proposed and compared several detection metrics for CVAE.
result CVAE-MSE and latent-based scores outperform ANMF-Tyler.
A new principle and method improve out-of-distribution detection in generative models.
problem Out-of-distribution detection in deep generative models often fails due to poor likelihood estimates.
method Introducing the Likelihood Path (LPath) principle and new theoretical tools for OOD detection.
result Non-asymptotic provable OOD detection guarantees for variational autoencoders (VAEs).
This work improves OOD detection using deep generative models by approximating Fisher information metrics.
problem Deep generative models often incorrectly infer higher likelihoods for out-of-distribution data.
method Approximating Fisher information metrics using gradient norms of data points.
result The method outperforms existing OOD detection techniques.
The Mahalanobis distance-based confidence score, a recently proposed anomaly detection method for pre-trained neural classifiers, achieves state-of-the-art performance on both out-of-distribution (OoD) and adversarial examples detection. This work analyzes why this method exhibits such strong performance in practical s…
OPNP prunes parameters and neurons to improve OOD detection without training.
problem Detecting out-of-distribution samples in real-world machine learning models.
method OPNP approach that identifies and removes sensitive parameters and neurons.
result OPNP consistently outperforms existing methods on multiple OOD detection tasks.
Paper proposes SCQ and P-TAMS for structured OOD testing in high-stakes ML.
problem Difficulty of incorporating auxiliary information in traditional conformal methods.
method Structure-adaptive conformal q-value (SCQ) and pseudo-score-guided transductive automated model selection (P-TAMS).
result Unified framework controls false discovery rate and improves power across diverse settings.
Improved anomaly detection using adversarial mirrored autoencoders.
problem Detecting out-of-distribution samples in machine learning.
method Adversarial Mirrored Autoencoder (AMA) with latent space regularization.
result AMA improves anomaly detection performance on OOD detection benchmarks.
ATOM improves robust OOD detection by mining informative auxiliary examples.
problem Robust OOD detection in open-world settings is challenging due to adversarial inputs.
method ATOM combines adversarial training with outlier mining to improve robustness.
result ATOM achieves state-of-the-art performance in OOD detection, reducing FPR by up to 57.99%.
Paper investigates learnability of OOD detection under various conditions.
problem Learnability of OOD detection under different scenarios.
method Investigates PAC learning theory, proves impossibility theorems, and provides necessary and sufficient conditions.
result Some conditions for learnability of OOD detection may not hold in practical scenarios.
Paper investigates OOD detection learnability under various conditions.
problem Learnability of OOD detection under diverse and unknown test data.
method PAC learning theory applied to OOD detection, proving impossibility theorems and necessary conditions.
result Some conditions for learnability hold in practical scenarios.
New method improves OOD detection without sacrificing generalization.
problem Improving OOD detection models that also generalize well.
method Decouples uncertainty learning from Bayesian perspective.
result Achieves state-of-the-art OOD detection performance.
PaRCE estimates model confidence for CNNs across various uncertainties.
problem Limited holistic approach to estimating perception model confidence in CNNs.
method Probabilistic and reconstruction-based competency estimation.
result PaRCE best distinguishes between various types of samples and regions.
LoD improves model safety by integrating unlabeled wild data, reducing OOD misclassification.
problem Improving model safety and reliability using unlabeled wild data containing both in-distribution and out-of-distribution samples.
method Intentionally label-noisifying unlabeled wild data to enable joint learning of labeled ID and OOD data, distinguishing losses between ID and OOD samples.
result LoD framework achieves superior OOD detection without requiring thresholds, improving model safety.
Paper presents a novel framework for OOD learning with human feedback.
problem Multifaceted challenges of OOD generalization and detection.
method Selective human feedback for labeling a small number of informative samples from wild data distribution.
result Enhanced robustness and precision in handling OOD scenarios.
The paper exposes common misconceptions about OOD detection and proposes a new framework.
problem Density-based OOD detection fails in deep learning settings.
method Proposes the OOD proxy framework to unify likelihood-ratio-based methods.
result Likelihood ratio is a principled method for OOD detection.
The paper explores how model complexity affects OOD detection performance.
problem Ensuring reliability and safety of machine learning systems through OOD detection.
method Investigates the relationship between model capacity and OOD detection performance using empirical and theoretical analysis.
result The Double Descent phenomenon is observed in post-hoc OOD detection, indicating that overparameterization can enhance OOD detection.