Probability Models for Open Set Recognition

By Walter J. Scheirer, Lalit P. Jain, Terrance E. Boult. 2014

661 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Media Technology, Probabilistic & Bayesian Methods

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Media Technology

In Probabilistic & Bayesian Methods

Citations in the corpus

Simple and Principled Uncertainty Estimation with Deterministic Deep Learning via Distance Awareness (2020, background): “As a result, the full data-generating distribution p∗(y|x) is in fact a mixture of an in-domain (IND) distribution pIND(y|x) = p∗(y|x,x ∈XIND) and also an OOD distribution pOOD(y|x) = p∗(y|x,x 6∈XIND) [52, 66]: p∗(y|x) = p∗(y,x ∈XIND|x) + p∗(y,x 6∈XIND|x) = p∗(y|x,x ∈XIND)∗ p∗(x ∈XIND)+ p∗(y|x,x 6∈X”
Model-Reuse Attacks on Deep Learning Systems (2018, background): “One may suggest using incremental learning [15, 48], which retrains an existing model to accommodate new data, or open-set learning, which extends a given model to new classes during inference [4, 51].”
Is Deep Learning Safe for Robot Vision? Adversarial Examples Against the iCub Humanoid (2017, methodology): “5-class classification (based on the combination of anomaly detectors and two-class classifiers) [3] to open-set recognition techniques [23, 2].”
iCaRL: Incremental Classifier and Representation Learning (2016, background): “rementally learn object classiﬁers. Royer and Lampert [25] adapt classiﬁers to a time-varying data stream but their method cannot handle newly appearing classes. Li and Wechsler [15], Scheirer et al. [29, 28], as well as Bendale and Boult [2] aimed at the related but distinct problem of Open Set Rec”
Few-Shot Continual Active Learning by a Robot (2022, methodology): “Another related field to active learning is open-set recognition (or out of the distribution detection (OOD)) [37, 38, 38, 39], in which a model that is trained on a set of tasks/classes might face samples belonging previously unknown classes in the test set.”
Towards Open World Object Detection (2021, background): “Follow up works [23, 59] extended the open set framework to multi-class classiﬁer setting with probabilistic models to account for the fading away classiﬁer conﬁdences in case of unknown classes.”
The limits and potentials of deep learning for robotics (2018, background): “These half-spaces are inﬁnite in extent, meaning unknown samples far from the support of known training data can receive a positive label [103].”
Towards Open World Recognition (2014, methodology): “η separately, we can apply Theorem 1 of [33] to each Mj,τj ,y(x) yielding a τj such that the function Mj,τj ,y(x) > 0 defines a labeled region lj(τj) ⊂ X with zero open space risk.”
A Wholistic View of Continual Learning with Deep Neural Networks: Forgotten Lessons and the Bridge to Active and Open World Learning (2020, background): “Scheirer et al. (2014) give an intuitive example based on support vector machines.”
Learning Cumulatively to Become More Knowledgeable (2016, methodology): “[32] introduced the Compact Abating Probability (CAP) model, which explained how thresholding the probabilistic output of RBF Oneclass SVM manages the open space risk.”
An Empirical Study and Analysis of Generalized Zero-Shot Learning for Object Recognition in the Wild (2016, background): “Finally, open set recognition [35,36,37] considers testing on both types of classes, but treating the unseen ones as a single outlier class.”
Recent Advances in Open Set Recognition: A Survey (2018, methodology): “Second, the reject thresholds are recommended to set according to the problem openness [22].”
Open Category Detection with PAC Guarantees (2018, methodology): “OCSVMs have been employed in a multi-class setting similar to open category detection (Heﬂin et al., 2012; Pritsos & Stamatatos, 2013).”
Secure Kernel Machines against Evasion Attacks (2016, background): “This has also been discussed in [28], exploiting a probabilistic model defined for open-set recognition, where the goal is to find enclosed decision functions around known training classes, to be able to detect novel classes at test time.”
Neural networks and neuroscience-inspired computer vision. (2014, background): “Indeed, even the venerable, largely solved MNIST hand-written digit recognition dataset [45] (current systems achieve in excess of 99% accuracy [46]) becomes difﬁcult when the system doesn’t have access during training to examples of all of the possible digits it might encounter during testing [47].”
The Extreme Value Machine (2015, methodology): “Nearest Neighbor (NN) classifiers with CAP probability estimation [12], and 1-vs-Rest RBF SVMs with Platt’s probability estimator [44] are also shown.”
Place categorization and semantic mapping on a mobile robot (2015, methodology): “Recently [7] used ConvNets to beat all competing approaches on the task of semantic place categorization.”
Unseen Class Discovery in Open-world Classification (2018, methodology): “Since many existing methods exist for (1) and (3) (Scheirer et al., 2013; 2014; Bendale & Boult, 2015; Fei & Liu, 2016; Bendale & Boult, 2016), we will not study them in this paper.”