Filters








36,078 Hits in 3.6 sec

Predicting with Confidence on Unseen Distributions [article]

Devin Guillory, Vaishaal Shankar, Sayna Ebrahimi, Trevor Darrell, Ludwig Schmidt
<span title="2021-08-19">2021</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
As a result, predicting model performance on unseen distributions is an important challenge.  ...  Our work connects techniques from domain adaptation and predictive uncertainty literature, and allows us to predict model accuracy on challenging unseen distributions without access to labeled data.  ...  materials, at predicting accuracy on natural and synthetic unseen distributions.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2107.03315v2">arXiv:2107.03315v2</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/ewlcswqpljfmjmsranrap6dkw4">fatcat:ewlcswqpljfmjmsranrap6dkw4</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20210911120634/https://arxiv.org/pdf/2107.03315v2.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/97/ed/97eddf3ef938d22684016f8fa8ec14c944be4972.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2107.03315v2" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Adaptive Confidence Smoothing for Generalized Zero-Shot Learning [article]

Yuval Atzmon, Gal Chechik
<span title="2019-10-07">2019</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
Thanks to COSMO's modular structure, instead of trying to perform well both on seen and on unseen classes, models can focus on accurate classification of unseen classes, and later consider seen class models  ...  We test our approach, adaptive confidence smoothing (COSMO), on four standard GZSL benchmark datasets and find that it largely outperforms state-of-the-art GZSL models.  ...  In Bayesian estimation, one combines the data (here, the predicted confidence) with a prior distribution (here, our prior belief).  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/1812.09903v3">arXiv:1812.09903v3</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/netlbpcgyrgh3ce7lpajpgpdim">fatcat:netlbpcgyrgh3ce7lpajpgpdim</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200908084023/https://arxiv.org/pdf/1812.09903v3.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/e4/59/e459697015465783ad7ec5d25cd8bcfa2fa1e25b.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/1812.09903v3" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Adaptive Confidence Smoothing for Generalized Zero-Shot Learning

Yuval Atzmon, Gal Chechik
<span title="">2019</span> <i title="IEEE"> <a target="_blank" rel="noopener" href="https://fatcat.wiki/container/ilwxppn4d5hizekyd3ndvy2mii" style="color: black;">2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)</a> </i> &nbsp;
We test our approach, adaptive confidence smoothing (COSMO ), on four standard GZSL benchmark datasets and find that it largely outperforms state-of-the-art GZSL models.  ...  We address two main difficulties in this approach: How to provide an accurate estimate of the gating probability without any training samples for unseen classes; and how to use expert predictions when  ...  Compared with Independent-Hard, COSMO shows a relative improvement from 58.3% to 63.6% on AWA, 35.1% to 41% on SUN and 44.6% to 50.2% on CUB.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener noreferrer" href="https://doi.org/10.1109/cvpr.2019.01194">doi:10.1109/cvpr.2019.01194</a> <a target="_blank" rel="external noopener" href="https://dblp.org/rec/conf/cvpr/AtzmonC19.html">dblp:conf/cvpr/AtzmonC19</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/njx24fnwe5eytksmtzwy3sjxwa">fatcat:njx24fnwe5eytksmtzwy3sjxwa</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20190611125406/http://openaccess.thecvf.com/content_CVPR_2019/papers/Atzmon_Adaptive_Confidence_Smoothing_for_Generalized_Zero-Shot_Learning_CVPR_2019_paper.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/90/bb/90bbacad4bf9084d7a76de019b99a4622c238e57.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener noreferrer" href="https://doi.org/10.1109/cvpr.2019.01194"> <button class="ui left aligned compact blue labeled icon button serp-button"> <i class="external alternate icon"></i> ieee.com </button> </a>

Domain segmentation and adjustment for generalized zero-shot learning [article]

Xinsheng Wang, Shanmin Pang, Jihua Zhu
<span title="2020-02-01">2020</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
Extensive experiments on five benchmark datasets show that the proposed method exhibits competitive performance compared with that based on generative models.  ...  In practice, we propose a threshold and probabilistic distribution joint method to segment the testing instances into seen, unseen and uncertain domains.  ...  The upper right is the distribution of confidence scores without TC, and the bottom right is that with TC.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2002.00226v1">arXiv:2002.00226v1</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/aipudnnmfre7nlyss5m2tef7va">fatcat:aipudnnmfre7nlyss5m2tef7va</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200322011733/https://arxiv.org/pdf/2002.00226v1.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2002.00226v1" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Probabilistic Object Classification using CNN ML-MAP layers [article]

G. Melotti, C. Premebida, J.J. Bird, D.R. Faria, N. Gonçalves
<span title="2020-08-24">2020</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
Experiments with calibrated and the proposed prediction layers are carried out on object classification using data from the KITTI database.  ...  To reduce the overconfidence without compromising the classification performance, we introduce a CNN probabilistic approach based on distributions calculated in the network's Logit layer.  ...  Fig. 4 : 4 Prediction scores, on the unseen data (comprising non-trained classes: 'person sit.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2005.14565v2">arXiv:2005.14565v2</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/fygoq5olvrbltdnzrwxlp2b75i">fatcat:fygoq5olvrbltdnzrwxlp2b75i</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200906233142/https://arxiv.org/pdf/2005.14565v2.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2005.14565v2" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Using UNSEEN trends to detect decadal changes in 100-year precipitation extremes

T. Kelder, M. Müller, L. J. Slater, T. I. Marjoribanks, R. L. Wilby, C. Prudhomme, P. Bohlinger, L. Ferranti, T. Nipen
<span title="2020-11-27">2020</span> <i title="Springer Science and Business Media LLC"> <a target="_blank" rel="noopener" href="https://fatcat.wiki/container/cmqjv25uprez3fmawzlapq4zzm" style="color: black;">npj Climate and Atmospheric Science</a> </i> &nbsp;
We fit the GEV distribution to the UNSEEN ensemble with a time covariate to facilitate detection of changes in 100-year precipitation values over a period of 35 years (1981–2015).  ...  To increase the historical record 100-fold, we apply the UNprecedented Simulated Extreme ENsemble (UNSEEN) approach, by pooling ensemble members and lead times from the ECMWF seasonal prediction system  ...  Hence, the trend estimates and confidence intervals in the 100-year return intervals expressed as UNSEEN distribution characteristics compared with the observed record.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener noreferrer" href="https://doi.org/10.1038/s41612-020-00149-4">doi:10.1038/s41612-020-00149-4</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/5uidor5ao5aczob2hepf4k6yhu">fatcat:5uidor5ao5aczob2hepf4k6yhu</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20210428035137/https://www.duo.uio.no/bitstream/handle/10852/83033/s41612-020-00149-4.pdf?sequence=2" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/4d/c1/4dc13ef8dc4215d22bb94a4edea3b4055284767f.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener noreferrer" href="https://doi.org/10.1038/s41612-020-00149-4"> <button class="ui left aligned compact blue labeled icon button serp-button"> <i class="unlock alternate icon" style="background-color: #fb971f;"></i> Publisher / doi.org </button> </a>

Reducing Overconfidence Predictions for Autonomous Driving Perception [article]

Gledson Melotti, Cristiano Premebida, Jordan J. Bird, Diego R. Faria, Nuno Gonçalves
<span title="2022-05-12">2022</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
Given this, the experiments in this work propose a probabilistic approach based on distributions calculated out of the Logit layer scores of pre-trained networks.  ...  the benefit of enabling interpretable probabilistic predictions.  ...  Additionally, given an object belonging to a non-trained/unseen class (e.g., an unexpected object on the road), how confident is the model's prediction?  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2202.07825v2">arXiv:2202.07825v2</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/cjh7upmqpveqdcohvvroinfvze">fatcat:cjh7upmqpveqdcohvvroinfvze</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20220518231142/https://arxiv.org/pdf/2202.07825v2.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/a7/b6/a7b6368c5070efc2a2a3f77f069a50f6bdfcff7c.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2202.07825v2" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Investigation of Uncertainty of Deep Learning-based Object Classification on Radar Spectra [article]

Kanil Patel, William Beluch, Kilian Rambach, Adriana-Eliza Cozma, Michael Pfeiffer, Bin Yang
<span title="2021-06-01">2021</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
We find that in agreement with phenomena observed in the literature,deep radar classifiers are overly confident, even in their wrong predictions.  ...  We show that by applying state-of-the-art post-hoc uncertainty calibration, the quality of confidence measures can be significantly improved,thereby partially resolving the over-confidence problem.  ...  on wrong, unseen and unknown data.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2106.05870v1">arXiv:2106.05870v1</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/wicivzdfhrarpf7a3okninems4">fatcat:wicivzdfhrarpf7a3okninems4</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20210612150802/https://arxiv.org/pdf/2106.05870v1.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/c5/03/c50327062419ebe5fb1eebe0ed49fba08210cb3e.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2106.05870v1" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Dont Even Look Once: Synthesizing Features for Zero-Shot Detection [article]

Pengkai Zhu, Hanxiao Wang, Venkatesh Saligrama
<span title="2020-04-10">2020</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
At a fundamental level, while vanilla detectors are capable of proposing bounding boxes, which include unseen objects, they are often incapable of assigning high-confidence to unseen objects, due to the  ...  data with ground truth bounding boxes is simply not scalable.  ...  We seek to improve confidence predictions on bounding boxes with sufficient overlap with seen and unseen objects, while still ensuring low confidence on bounding boxes that primarily contain background  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/1911.07933v3">arXiv:1911.07933v3</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/c4nesge7bzfknm55zkzs66ywp4">fatcat:c4nesge7bzfknm55zkzs66ywp4</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200414221541/https://arxiv.org/pdf/1911.07933v3.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/1911.07933v3" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Prediction Confidence from Neighbors [article]

Mark Philip Philipsen, Thomas Baltzer Moeslund
<span title="2020-03-31">2020</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
The distance between unseen samples and nearby training samples proves to be correlated to the prediction error of unseen samples.  ...  Depending on the acceptable degree of error, predictions can either be trusted or rejected based on the distance to training samples.  ...  The contributions can be summarized as: 1) Confidence measure for unseen samples based on distance to training set neighbours in feature space. II.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2003.14047v1">arXiv:2003.14047v1</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/nlis26s72jhnbozp6ftnmaf55q">fatcat:nlis26s72jhnbozp6ftnmaf55q</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200402045831/https://arxiv.org/pdf/2003.14047v1.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/52/23/5223ed2abdde925a044763a1a31916905c5029c4.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2003.14047v1" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Toward Reliable Models for Authenticating Multimedia Content: Detecting Resampling Artifacts With Bayesian Neural Networks [article]

Anatol Maier, Benedikt Lorch, Christian Riess
<span title="2020-07-28">2020</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
In this work, we make a first step toward redesigning forensic algorithms with a strong focus on reliability.  ...  However, most existing methods are challenged by out-of-distribution data, i.e., with characteristics that are not covered in the training set.  ...  As prior distribution we assume a zero-mean Gaussian distribution with unit-variance. For inference, we use Eqn. 6 with n = 50 Monte Carlo draws, and we calculate the predictive variance via Eqn. 7.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2007.14132v1">arXiv:2007.14132v1</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/drmi5mcizzdpvpldtlud7fcxnm">fatcat:drmi5mcizzdpvpldtlud7fcxnm</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200730221138/https://arxiv.org/pdf/2007.14132v1.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2007.14132v1" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Why is the Mahalanobis Distance Effective for Anomaly Detection? [article]

Ryo Kamoi, Kei Kobayashi
<span title="2020-04-30">2020</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
This suggests that the reason the Mahalanobis confidence score works so well is mistaken, and makes use of different information from ODIN, another popular OoD detection method based on prediction confidence  ...  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)  ...  Contribution The Mahalanobis confidence score [20] had been believed to function based on the prediction confidence of classifiers.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2003.00402v2">arXiv:2003.00402v2</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/jeackg3wxbhqpa7ew6h2uwh5bm">fatcat:jeackg3wxbhqpa7ew6h2uwh5bm</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200506190735/https://arxiv.org/pdf/2003.00402v2.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2003.00402v2" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Revisiting Explicit Regularization in Neural Networks for Well-Calibrated Predictive Uncertainty [article]

Taejong Joo, Uijung Chung
<span title="2021-02-06">2021</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
We then explore explicit regularization techniques for improving the log-likelihood on unseen samples, which provides well-calibrated predictive uncertainty.  ...  However, the impressive generalization performance of neural networks with only implicit regularization may be at odds with this conventional wisdom.  ...  That is, restricting the predictive confidence φ W mx (x) on training samples will directly impact the predictive confidence on unseen samples and therefore can improve the reliability of the predictive  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2006.06399v3">arXiv:2006.06399v3</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/whuguazt2febfcoq6zxd2wkfxu">fatcat:whuguazt2febfcoq6zxd2wkfxu</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20201006012703/https://arxiv.org/pdf/2006.06399v2.pdf" title="fulltext PDF download [not primary version]" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <span style="color: #f43e3e;">&#10033;</span> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/2006.06399v3" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Open Set Recognition Through Deep Neural Network Uncertainty: Does Out-of-Distribution Detection Require Generative Classifiers? [article]

Martin Mundt, Iuliia Pliushch, Sagnik Majumder, Visvanathan Ramesh
<span title="2019-08-26">2019</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
We present an analysis of predictive uncertainty based out-of-distribution detection for different approaches to estimate various models' epistemic uncertainty and contrast it with extreme value theory  ...  While the former alone does not seem to be enough to overcome this challenge, we demonstrate that uncertainty goes hand in hand with the latter method.  ...  Neither of the approaches is able to avoid over-confident predictions on previously unseen datasets, even if MCD fares much better in separating the distributions.  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/1908.09625v1">arXiv:1908.09625v1</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/afpajbsf3nff5afvx6bvdhgrju">fatcat:afpajbsf3nff5afvx6bvdhgrju</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200907032452/https://arxiv.org/pdf/1908.09625v1.pdf" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/dd/b9/ddb9959381a2c68fad70e4660d1e9f1f1eece879.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/1908.09625v1" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>

Confidence-Calibrated Adversarial Training: Generalizing to Unseen Attacks [article]

David Stutz, Matthias Hein, Bernt Schiele
<span title="2020-06-30">2020</span> <i > arXiv </i> &nbsp; <span class="release-stage" >pre-print</span>
Our confidence-calibrated adversarial training (CCAT) tackles this problem by biasing the model towards low confidence predictions on adversarial examples.  ...  By allowing to reject examples with low confidence, robustness generalizes beyond the threat model employed during training.  ...  During training, we train the network to predict a convex combination of (correct) one-hot distribution on clean examples and uniform distribution on adversarial examples as target distribution within  ... 
<span class="external-identifiers"> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/1910.06259v4">arXiv:1910.06259v4</a> <a target="_blank" rel="external noopener" href="https://fatcat.wiki/release/hycho5barvhxtpifgkv73uo42i">fatcat:hycho5barvhxtpifgkv73uo42i</a> </span>
<a target="_blank" rel="noopener" href="https://web.archive.org/web/20200706170114/https://arxiv.org/pdf/1910.06259v4.pdf" title="fulltext PDF download [not primary version]" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <span style="color: #f43e3e;">&#10033;</span> <div class="menu fulltext-thumbnail"> <img src="https://blobs.fatcat.wiki/thumbnail/pdf/4e/ad/4ead42585659cf23394786ccd9fa1f203912cd6e.180px.jpg" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener" href="https://arxiv.org/abs/1910.06259v4" title="arxiv.org access"> <button class="ui compact blue labeled icon button serp-button"> <i class="file alternate outline icon"></i> arxiv.org </button> </a>
&laquo; Previous Showing results 1 &mdash; 15 out of 36,078 results