2,083 Hits in 2.9 sec

Deep Networks Provably Classify Data on Curves [article]

Tingran Wang, Sam Buchanan, Dar Gilboa, John Wright
2021 arXiv   pre-print
We study a model problem with such structure -- a binary classification task that uses a deep fully-connected neural network to classify data drawn from two disjoint smooth curves on the unit sphere.  ...  To our knowledge, this is the first generalization guarantee for deep networks with nonlinear data that depends only on intrinsic data properties.  ...  We thank Alberto Bietti for bringing to our attention relevant prior art on kernel regression on manifolds.  ... 
arXiv:2107.14324v2 fatcat:sz76vjvul5dhdemlsbjzdnarhy

Provable defenses against adversarial examples via the convex outer adversarial polytope [article]

Eric Wong, J. Zico Kolter
2018 arXiv   pre-print
We propose a method to learn deep ReLU-based classifiers that are provably robust against norm-bounded adversarial perturbations on the training data.  ...  We illustrate the approach on a number of tasks to train classifiers with robust adversarial guarantees (e.g. for MNIST, we produce a convolutional classifier that provably has less than 5.8 adversarial  ...  Schmidt for providing helpful comments on an earlier draft of this work.  ... 
arXiv:1711.00851v3 fatcat:u6dxtu4rtjg6rlywbtvqtcwe2u

Deep AUC Maximization for Medical Image Classification: Challenges and Opportunities [article]

Tianbao Yang
2021 arXiv   pre-print
Since AUC (aka area under ROC curve) is a standard performance measure for medical image classification, hence directly optimizing AUC could achieve a better performance for learning a deep neural network  ...  In this paper, we will discuss these recent results by highlighting (i) the advancements brought by stochastic non-convex optimization algorithms for DAM; (ii) the promising results on various medical  ...  network and to learn the classifier by DAM [59] .  ... 
arXiv:2111.02400v1 fatcat:4thbkrbbqndqbk3ksaiwc3zhzu

MACER: Attack-free and Scalable Robust Training via Maximizing Certified Radius [article]

Runtian Zhai, Chen Dan, Di He, Huan Zhang, Boqing Gong, Pradeep Ravikumar, Cho-Jui Hsieh, Liwei Wang
2020 arXiv   pre-print
In our experiments, we show that our method can be applied to modern deep neural networks on a wide range of datasets, including Cifar-10, ImageNet, MNIST, and SVHN.  ...  Recent work shows that randomized smoothing can be used to provide a certified l2 radius to smoothed classifiers, and our algorithm trains provably robust smoothed classifiers via MAximizing the CErtified  ...  Unfortunately, computing the robust radius (1) of a classifier induced by a deep neural network is very difficult. showed that computing the l 1 robust radius of a deep neural network is NP-hard.  ... 
arXiv:2001.02378v3 fatcat:zb2i5l4jt5gtxdytzn5ybrfm7u

Provable Guarantees for Understanding Out-of-Distribution Detection

Peyman Morteza, Yixuan Li
In particular, on CIFAR-100 as in-distribution data, our method outperforms a competitive baseline by 16.57% (FPR95).  ...  Lastly, we formally provide provable guarantees and comprehensive analysis of our method, underpinning how various properties of data distribution affect the performance of OOD detection.  ...  In Section 3, we extend GEM to deep neural networks and perform experiments on common benchmarks.  ... 
doi:10.1609/aaai.v36i7.20752 fatcat:bnuoatvokfggfb53dve46gz32m

Denoised Smoothing: A Provable Defense for Pretrained Classifiers [article]

Hadi Salman, Mingjie Sun, Greg Yang, Ashish Kapoor, J. Zico Kolter
2020 arXiv   pre-print
We present a method for provably defending any pretrained image classifier against ℓ_p adversarial attacks.  ...  This method, for instance, allows public vision API providers and users to seamlessly convert pretrained non-robust classification services into provably robust ones.  ...  When f is a deep neural network, computing p A and p B accurately is not practical.  ... 
arXiv:2003.01908v2 fatcat:i2h3ujuhjjab5azoruo2jibdce

Provable Guarantees for Understanding Out-of-distribution Detection [article]

Peyman Morteza, Yixuan Li
2021 arXiv   pre-print
In particular, on CIFAR-100 as in-distribution data, our method outperforms a competitive baseline by 16.57% (FPR95).  ...  Lastly, we formally provide provable guarantees and comprehensive analysis of our method, underpinning how various properties of data distribution affect the performance of OOD detection.  ...  In Section 3, we extend GEM to deep neural networks and perform experiments on common benchmarks.  ... 
arXiv:2112.00787v1 fatcat:3jilc2nnwzexhlqrjnkypjltfm

Deep Partition Aggregation: Provable Defense against General Poisoning Attacks [article]

Alexander Levine, Soheil Feizi
2021 arXiv   pre-print
Adversarial poisoning attacks distort training data in order to corrupt the test-time behavior of a classifier.  ...  We propose two novel provable defenses against poisoning attacks: (i) Deep Partition Aggregation (DPA), a certified defense against a general poisoning threat model, defined as the insertion or deletion  ...  DPA PRACTICAL IMPLEMENTATION DETAILS One of the advantages of DPA is that we can use deep neural networks for the base classifier f .  ... 
arXiv:2006.14768v2 fatcat:gfttpdivavgyvj3r3y5ahuercu

Provably Robust Deep Learning via Adversarially Trained Smoothed Classifiers [article]

Hadi Salman, Greg Yang, Jerry Li, Pengchuan Zhang, Huan Zhang, Ilya Razenshteyn, Sebastien Bubeck
2020 arXiv   pre-print
Recent works have shown the effectiveness of randomized smoothing as a scalable technique for building neural network-based classifiers that are provably robust to ℓ_2-norm adversarial perturbations.  ...  We demonstrate through extensive experimentation that our method consistently outperforms all existing provably ℓ_2-robust classifiers by a significant margin on ImageNet and CIFAR-10, establishing the  ...  More Data for Better Provable Robustness We explore using more data to improve the robustness of smoothed classifiers.  ... 
arXiv:1906.04584v5 fatcat:whnwu6rcqvf33fuxolghsqsf3u

Scaling provable adversarial defenses [article]

Eric Wong, Frank R. Schmidt, Jan Hendrik Metzen, J. Zico Kolter
2018 arXiv   pre-print
On both MNIST and CIFAR data sets, we train classifiers that improve substantially on the state of the art in provable robust adversarial error bounds: from 5.8 (with ℓ_∞ perturbations of ϵ=0.1), and from  ...  Recent work has developed methods for learning deep network classifiers that are provably robust to norm-bounded adversarial perturbation; however, these methods are currently only possible for relatively  ...  constructing provably robust bounds for general deep network architectures, using Fenchel duality.  ... 
arXiv:1805.12514v2 fatcat:3ctlp2ltxjc7vpelhbfzkdlea4

Towards neural networks that provably know when they don't know [article]

Alexander Meinke, Matthias Hein
2020 arXiv   pre-print
Our approach can be used with ReLU networks and provides provably low confidence predictions far away from the training data as well as the first certificates for low confidence predictions in a neighborhood  ...  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.  ...  The first one says that the classifier has provably low confidence far away from the training data, where an explicit bound on the minimal distance is provided, and the second provides an upper bound on  ... 
arXiv:1909.12180v2 fatcat:ieybxvbtezeoxi32ehtdr76sua

Uncertainty Calibration for Deep Audio Classifiers [article]

Tong Ye, Shijing Si, Jianzong Wang, Ning Cheng, Jing Xiao
2022 arXiv   pre-print
Results indicate that uncalibrated deep audio classifiers may be over-confident, and SNGP performs the best and is very efficient on the two datasets of this paper.  ...  In this work, we investigate the uncertainty calibration for deep audio classifiers.  ...  In these experiments, a neural network is first trained on some ESC-50 data, which represents the in-distribution examples.  ... 
arXiv:2206.13071v1 fatcat:uuvfzpkgvzfxfbol2lsyahcpmm

Development of an Automated Screening System for Retinopathy of Prematurity Using a Deep Neural Network for Wide-Angle Retinal Images

Yinsheng Zhang, Li Wang, Zhenquan Wu, Jian Zeng, Yi Chen, Ruyin Tian, Jinfeng Zhao, Guoming Zhang
2019 IEEE Access  
The objective of this paper is to evaluate the performance of a deep neural network (DNN) for the automated screening of ROP.  ...  The receiver operating characteristic (ROC) curve, ROC area under the curve, and precision-recall (P-R) curve on the test dataset were analyzed.  ...  TABLE 4 . 4 Ophthalmologists and deep neural network performances on the test set.  ... 
doi:10.1109/access.2018.2881042 fatcat:defwis45bvhjjb7bupfhbzplia

Overfitting in adversarially robust deep learning [article]

Leslie Rice, Eric Wong, J. Zico Kolter
2020 arXiv   pre-print
Finally, we study several classical and modern deep learning remedies for overfitting, including regularization and data augmentation, and find that no approach in isolation improves significantly upon  ...  It is common practice in deep learning to use overparameterized networks and train for as long as possible; there are numerous studies that show, both theoretically and empirically, that such practices  ...  the network on adversarially perturbed inputs instead of on clean data (Goodfellow et al., 2014) .  ... 
arXiv:2002.11569v2 fatcat:aa6lwu5fi5ab7h2qgr4sqpfp3i

Deep-learnt classification of light curves

A Mahabal, K Sheth, F Gieseke, A Pai, S G Djorgovski, A J Drake, M J Graham
2017 2017 IEEE Symposium Series on Computational Intelligence (SSCI)  
In this work, we transform the time series to two-dimensional light curve representations in order to classify them using modern deep learning techniques.  ...  In particular, we show that convolutional neural networks based classifiers work well for broad characterization and classification.  ...  of the deep network.  ... 
doi:10.1109/ssci.2017.8280984 dblp:conf/ssci/MahabalSGPDDG17 fatcat:xnftyt2savdi3ihveovovw2vgu
« Previous Showing results 1 — 15 out of 2,083 results