Filters








24,498 Hits in 2.4 sec

Boosting Black Box Variational Inference [article]

Francesco Locatello, Gideon Dresdner, Rajiv Khanna, Isabel Valera, and Gunnar Rätsch
2018 arXiv   pre-print
These theoretical enhancements allow for black box implementation of the boosting subroutine.  ...  Variational Inference (VI) is a popular technique that achieves tractability by choosing a relatively simple variational family.  ...  box solvers (e.g. black box variational inference [19] ). • We propose a novel stopping criterion using the duality gap from FW, which is applicable to any boosting VI algorithm.  ... 
arXiv:1806.02185v5 fatcat:upsqt5t3wjfijewd5763tlnrr4

Variational Boosting: Iteratively Refining Posterior Approximations [article]

Andrew C. Miller, Nicholas Foti, Ryan P. Adams
2017 arXiv   pre-print
We propose a black-box variational inference method to approximate intractable distributions with an increasingly rich approximating class.  ...  We apply variational boosting to synthetic and real statistical models, and show that resulting posterior inferences compare favorably to existing posterior approximation algorithms in both accuracy and  ...  Nonparametric variational inference [9] is a black-box variational inference algorithm that approximates a target distribution with a mixture of equally-weighted isotropic normals.  ... 
arXiv:1611.06585v2 fatcat:3nikxq4acjhmjm4yvpyv2ps63a

Boosting Variational Inference With Locally Adaptive Step-Sizes [article]

Gideon Dresdner, Saurav Shekhar, Fabian Pedregosa, Francesco Locatello, Gunnar Rätsch
2021 arXiv   pre-print
The main obstacle to widespread adoption of Boosting Variational Inference is the amount of resources necessary to improve over a strong Variational Inference baseline.  ...  Instead, Boosting Variational Inference allows practitioners to obtain increasingly good posterior approximations by spending more compute.  ...  Boosting Variational Inference Boosting Variational Inference (VI) aims to solve an expanded version of the problem defined in Equation (1) by optimizing over the convex hull of Q defined as, conv(Q) def  ... 
arXiv:2105.09240v1 fatcat:7w7qpa3pt5gfhnwcvh3prgmduu

Assessment of Validity Conditions for Black-Box EMI Modelling of DC/DC Converters

Lu Wan, Abduselam Beshir, Xinglong Wu, Xiaokang Liu, Flavia Grassi, Giordano Spadacini, Sergio Pignari
2021 2021 IEEE International Joint EMC/SI/PI and EMC Europe Symposium  
In order to investigate suitable conditions assuring effectiveness of black-box modelling for a boost converter, this work investigates the role that the converter input capacitors and the functional inductor  ...  play in masking the inherent non-linear and time variant behavior of the switching modules, and their impact on the effectiveness of the proposed black-box model.  ...  Assessment of Validity Conditions for Black-Box EMI Modelling of DC/DC Converters  ... 
doi:10.1109/emc/si/pi/emceurope52599.2021.9559274 fatcat:ntpif74nnvhbhpmhrskwhpg34u

Cognitive Deficit of Deep Learning in Numerosity [article]

Xiaolin Wu, Xi Zhang, Xiao Shu
2018 arXiv   pre-print
But somewhat disappointingly, extensive experiments of the type of cognitive psychology demonstrate that the examples-driven black box DL cannot see through superficial variations in visual representations  ...  Also, we investigate, using subitizing as a test bed, the ways to aid the black box DL by cognitive priors derived from human insight.  ...  If the black-box DL approach is ever able to attain the level of human intelligence, then it has to generalize or reason beyond statistical inference under the i.i.d. condition.  ... 
arXiv:1802.05160v4 fatcat:iajgpajfd5hffdnuauauzxnma4

Cognitive Deficit of Deep Learning in Numerosity

Xiaolin Wu, Xi Zhang, Xiao Shu
2019 PROCEEDINGS OF THE THIRTIETH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE AND THE TWENTY-EIGHTH INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE CONFERENCE  
But somewhat disappointingly, extensive experiments of the type of cognitive psychology demonstrate that the examples-driven black box DL cannot see through superficial variations in visual representations  ...  Also, we investigate, using subitizing as a test bed, the ways to aid the black box DL by cognitive priors derived from human insight.  ...  If the black-box DL approach is ever able to attain the level of human intelligence, then it has to generalize or reason beyond statistical inference under the i.i.d. condition.  ... 
doi:10.1609/aaai.v33i01.33011303 fatcat:mwmsanc2djfmvpg7tg4a7aagvm

Automated versus do-it-yourself methods for causal inference: Lessons learned from a data analysis competition [article]

Vincent Dorie, Jennifer Hill, Uri Shalit, Marc Scott, Dan Cervone
2018 arXiv   pre-print
Results from 30 competitors across the two versions of the competition (black box algorithms and do-it-yourself analyses) are presented along with post-hoc analyses that reveal information about the characteristics  ...  of causal inference strategies and settings that affect performance.  ...  CAUSAL INFERENCE SUBMISSIONS AND KEY FEATURES We received 15 submissions for the DIY portion of competition and 15 for the black-box section.  ... 
arXiv:1707.02641v5 fatcat:pvfnziorg5fgjasp767uupiutu

Stealing Neural Networks via Timing Side Channels [article]

Vasisht Duddu, Debasis Samanta, D Vijay Rao, Valentina E. Balas
2019 arXiv   pre-print
In this paper, a black box Neural Network extraction attack is proposed by exploiting the timing side channels to infer the depth of the network.  ...  Here, an adversary can extract the Neural Network parameters, infer the regularization hyperparameter, identify if a data point was part of the training data, and generate effective transferable adversarial  ...  A major security question in such a black box setting like MLaaS addressed in this paper is whether a weak adversary in a black box setting can efficiently infer target Neural Network attributes by exploiting  ... 
arXiv:1812.11720v4 fatcat:hts4m64pabh37fp2jalgkoysu4

LOGAN: Membership Inference Attacks Against Generative Models

Jamie Hayes, Luca Melis, George Danezis, Emiliano De Cristofaro
2019 Proceedings on Privacy Enhancing Technologies  
We present attacks based on both white-box and black-box access to the target model, against several state-of-the-art generative models, over datasets of complex representations of faces (LFW), objects  ...  In this paper, we present the first membership inference attacks against generative models: given a data point, the adversary determines whether or not it was used to train the model.  ...  We devise a white-box attack that is an excellent indicator of overfitting in generative models, and a black-box attack that can be mounted through Generative Adversarial Networks, and show how to boost  ... 
doi:10.2478/popets-2019-0008 dblp:journals/popets/HayesMDC19 fatcat:2x3xnelx3zf5tgkkufvgvubzzq

LOGAN: Membership Inference Attacks Against Generative Models [article]

Jamie Hayes, Luca Melis, George Danezis, Emiliano De Cristofaro
2018 arXiv   pre-print
We present attacks based on both white-box and black-box access to the target model, against several state-of-the-art generative models, over datasets of complex representations of faces (LFW), objects  ...  In this paper, we present the first membership inference attacks against generative models: given a data point, the adversary determines whether or not it was used to train the model.  ...  boost the performance of 1 https://neuromation.io the black-box attack via auxiliary attacker knowledge of training/testing set; 3.  ... 
arXiv:1705.07663v4 fatcat:amddvcw7i5gf5jeh6wzisgavd4

Guess First to Enable Better Compression and Adversarial Robustness [article]

Sicheng Zhu, Bang An, Shiyu Niu
2020 arXiv   pre-print
In this paper, we try to leverage one of the mechanisms in human recognition and propose a bio-inspired classification framework in which model inference is conditioned on label hypothesis.  ...  We provide a class of training objectives for this framework and an information bottleneck regularizer which utilizes the advantage that label information can be discarded during inference.  ...  'b' denotes black-box attack and 'w' denotes white-box attack.  ... 
arXiv:2001.03311v1 fatcat:aiasa4uv3befbi2fejl7x3gb3e

Improving the robustness and accuracy of biomedical language models through adversarial training [article]

Milad Moradi, Matthias Samwald
2021 arXiv   pre-print
In addition, the models performance on clean data increased in average by 2.4 absolute present, demonstrating that adversarial training can boost generalization abilities of biomedical NLP systems.  ...  HotFlip and the white-box variation of TextBugger, generally caused larger declines in performance scores compared to the black-box attackers.  ...  TextBugger [26] can operate in both white-box and black-box attack scenarios.  ... 
arXiv:2111.08529v1 fatcat:f33bi5bfwrfgradm3tw53gt4ti

AID-Purifier: A Light Auxiliary Network for Boosting Adversarial Defense [article]

Duhun Hwang, Eunjung Lee, Wonjong Rhee
2021 arXiv   pre-print
We propose an AID-purifier that can boost the robustness of adversarially-trained networks by purifying their inputs.  ...  Black-box attack We generate black-box adversarial examples using a pre-trained VGG19 network (Simonyan & Zisserman, 2014) .  ...  PGD, C&W, DF, and MIM attacks are used to evaluate black-box robustness on SVHN. We report only the worst black-box accuracy in Table 10 . For the case of Mardy, a large improvement is achieved. G.  ... 
arXiv:2107.06456v1 fatcat:xngylhpldbaqrnt7cak2sy5uhy

GAN-Leaks: A Taxonomy of Membership Inference Attacks against GANs [article]

Dingfan Chen, Ning Yu, Yang Zhang, Mario Fritz
2019 arXiv   pre-print
Specifically, we present the first taxonomy of membership inference attacks, which encompasses not only existing attacks but also our novel ones.  ...  In this paper, we focus on membership inference attack against GANs that has the potential to reveal information about victim models' training data.  ...  Variational AutoEncoder (VAE).  ... 
arXiv:1909.03935v1 fatcat:ainfmkkbkbezhamjmbzmvhsw2m

Incorporating Scene Context and Object Layout into Appearance Modeling

Hamid Izadinia, Fereshteh Sadeghi, Ali Farhadi
2014 2014 IEEE Conference on Computer Vision and Pattern Recognition  
Our scene structure provides a level of scene understanding that is amenable to deep visual inferences.  ...  The Black Box Test Our scene structures provide a level of scene understanding that enables deep inferences such as the one used in Black Box Test.  ...  Figure 6 . 6 Black Box Test: What is behind the black box?  ... 
doi:10.1109/cvpr.2014.37 dblp:conf/cvpr/IzadiniaSF14 fatcat:f7d4j3okbffdrbinruo3yurxpe
« Previous Showing results 1 — 15 out of 24,498 results