Entropic Out-of-Distribution Detection: Seamless Detection of Unknown Examples

Macêdo, David; Ren, Tsang Ing; Zanchettin, Cleber; Oliveira, Adriano L. I.; Ludermir, Teresa Bernarda

doi:10.1109/tnnls.2021.3112897

Cited by 11 publications

(6 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A widely used baseline expresses the OOD score for an image

directly from discriminative predictions as

[ 40 ], where

computes dense logits based on the input image. Entropy-based detectors can deliver a similar performance [ 41 , 42 ]. Another line of work improves upon these baselines by pre-processing the input with anti-adversarial perturbations [ 32 ], which cause significant computational overhead.…”

Section: Related Workmentioning

confidence: 99%

Dense Out-of-Distribution Detection by Robust Learning on Synthetic Negative Data

Grcić,

Bevandić,

Kalafatić

et al. 2024

Sensors

View full text Add to dashboard Cite

Standard machine learning is unable to accommodate inputs which do not belong to the training distribution. The resulting models often give rise to confident incorrect predictions which may lead to devastating consequences. This problem is especially demanding in the context of dense prediction since input images may be only partially anomalous. Previous work has addressed dense out-of-distribution detection by discriminative training with respect to off-the-shelf negative datasets. However, real negative data may lead to over-optimistic evaluation due to possible overlap with test anomalies. To this end, we extend this approach by generating synthetic negative patches along the border of the inlier manifold. We leverage a jointly trained normalizing flow due to a coverage-oriented learning objective and the capability to generate samples at different resolutions. We detect anomalies according to a principled information-theoretic criterion which can be consistently applied through training and inference. The resulting models set the new state of the art on benchmarks for out-of-distribution detection in road-driving scenes and remote sensing imagery despite minimal computational overhead.

show abstract

“…A widely used baseline expresses the OOD score for an image

directly from discriminative predictions as

[ 40 ], where

Section: Related Workmentioning

confidence: 99%

Dense Out-of-Distribution Detection by Robust Learning on Synthetic Negative Data

Grcić,

Bevandić,

Kalafatić

et al. 2024

Sensors

View full text Add to dashboard Cite

show abstract

“…The first system was adapted from the learning confidence out-of-distribution detection model [24] which estimates learning confidence for neural networks and produces intuitively interpretable outputs. The other approach incorporates an entropic out-of-distribution detection model [40] which replaces the SoftMax loss with a novel loss function dealing with the weaknesses of SoftMax loss anisotropy, and tends to produce low entropy probability distributions which breaks the principle of maximum entropy. In OOD detection problems, only ID data is taken as training and validation dataset and combined ID and OOD data is taken as testing dataset.…”

Section: Out-of-distribution Detectionmentioning

confidence: 99%

“…al. in [40] solves the SoftMax loss drawbacks by replacing it with the Isotropy Maximization (IsoMax) loss. The proposed IsoMax loss is isotropic and follows the maximum entropy principle.…”

Section: ) Confidence-based Approachmentioning

confidence: 99%

“…The final models from each run were selected by measuring the classification accuracy of the ID task on the validation set since the OOD data is assumed to be unknown during training and validation. For the OOD detection task, we use the following rules to set any low confidence predictions as a negative detection (i.e., we prioritize OOD task AUROC value but other rules can be selected): (1) For the Entropy-based model, similar to the entropy-based model for comparison purpose, OOD samples are separated by a distance threshold which produces the highest OOD task AUROC on the test dataset [40]; (2) For the Confidence-based model, we selected a confidence threshold giving the highest OOD task AUROC on the test dataset. The results for detection error, F1, precision, and recall come from detection with these threshold selection rules.…”

Section: Ood Detection Performance For Different Sampling Ratesmentioning

confidence: 99%

See 1 more Smart Citation

Robust Cough Detection With Out-of-Distribution Detection

Chen

Attri

Barahona

et al. 2023

IEEE J. Biomed. Health Inform.

View full text Add to dashboard Cite

Cough is an important defense mechanism of the respiratory system and is also a symptom of lung diseases, such as asthma. Acoustic cough detection collected by portable recording devices is a convenient way to track potential condition worsening for patients who have asthma. However, the data used in building current cough detection models are often clean, containing a limited set of sound categories, and thus perform poorly when they are exposed to a variety of real-world sounds which could be picked up by portable recording devices. The sounds that are not learned by the model are referred to as Out-of-Distribution (OOD) data. In this work, we propose two robust cough detection methods combined with an OOD detection module, that removes OOD data without sacrificing the cough detection performance of the original system. These methods include adding a learning confidence parameter and maximizing entropy loss. Our experiments show that 1) the OOD system can produce dependable In-Distribution (ID) and OOD results at a sampling rate above 750 Hz; 2) the OOD sample detection tends to perform better for larger audio window sizes; 3) the model's overall accuracy and precision get better as the proportion of OOD samples increase in the acoustic signals; 4) a higher percentage of OOD data is needed to realize performance gains at lower sampling rates. The incorporation of OOD detection techniques improves cough detection performance by a significant margin and provides a valuable solution to real-world acoustic cough detection problems.

show abstract

“…This approach only obtained a 63% detection accuracy on the image data, which is not desirable for security critical applications (5) . Another method of thresholding the distance from a Gaussian distribution fitted to the target class representations, the author of (6) investigates a technique for identifying adversarial samples and states that the Mahalanobis distance detecting technique is the most vulnerable to attack. Fabio Carrara and fellow authors have proposed ENAD, an ensemble approach for adversarial detection that improves performance by integrating layer-specific scores from three independent detectors (LID, Mahalanobis, and OCSVM), achieving significantly enhanced performance on benchmark datasets, methods, and attacks but requiring training (7) .…”

Section: Introductionmentioning

confidence: 99%

Mitigating Gradient-Based Data Poisoning Attacks on Machine Learning Models: A Statistical Detection Method

Sanapala,

Gondi

2024

IJST

View full text Add to dashboard Cite

Objectives: This research paper aims to develop a novel method for identifying gradient-based data poisoning attacks on industrial applications like autonomous vehicles and intelligent healthcare systems relying on machine learning and deep learning techniques. These algorithms performs well only if they are trained on good quality dataset. However, the ML models are prone to data poisoning attacks, targeting the training dataset, manipulate its input samples such that the machine learning algorithm gets confused and produces wrong predictions. The current detection techniques are effective to detect known attacks and lack generalized detection to unknown attacks. To address this issue, this paper aims to integrate security elements within the machine learning framework, guaranteeing effective identification and mitigation of known and unknown threats and achieve generalized detection. Methods: ML Filter, a unique attack detection approach integrates ML-Filter Detection Algorithm and the Statistical Perturbation Bounds Identification Algorithm to determine the given dataset is poisoned or not. DBSCAN algorithm is used to divide the dataset into several smaller subsets and perform algorithmic analysis for detection. The performance of the proposed method is evaluated in terms of True positive rate and significance test accuracy. Findings: The probability distribution differences between original and poisoned datasets vary with change in perturbation size rather than the datasets and ML models use for application. This finding lead to determine the perturbation bounds using statistical pairwise distance metrics and corresponding significance tests computed on the results. ML Filter demonstrates a high detection rate of 99.63% for known attacks and achieves a generalized detection accuracy of 98% for unknown attacks. Novelty: A secured ML architecture and a unique statistical detection approach ML-Filter, effectively detect data poisoning attacks, demonstrating significant advancements in detecting both known and unknown threats in industrial applications utilizing machine learning and deep learning algorithms. Keywords: Privacy and security, Adversarial machine learning, Secured ML Architecture, ML-Filter, Statistical Perturbation Bounds Identification Algorithm

show abstract

Entropic Out-of-Distribution Detection: Seamless Detection of Unknown Examples

Cited by 11 publications

References 32 publications

Dense Out-of-Distribution Detection by Robust Learning on Synthetic Negative Data

Dense Out-of-Distribution Detection by Robust Learning on Synthetic Negative Data

Robust Cough Detection With Out-of-Distribution Detection

Mitigating Gradient-Based Data Poisoning Attacks on Machine Learning Models: A Statistical Detection Method

Contact Info

Product

Resources

About