Filters








122,731 Hits in 5.6 sec

A Witness Two-Sample Test [article]

Jonas M. Kübler, Wittawat Jitkrittum, Bernhard Schölkopf, Krikamol Muandet
2022 arXiv   pre-print
The Maximum Mean Discrepancy (MMD) has been the state-of-the-art nonparametric test for tackling the two-sample problem.  ...  We show that 1) the new test is consistent and has a well-controlled type-I error; 2) the optimal witness function is given by a precision-weighted mean in the reproducing kernel Hilbert space associated  ...  Benchmark CONCLUSION We introduced a principled approach to learn optimal witness functions for two-sample testing.  ... 
arXiv:2102.05573v3 fatcat:arosby3ihvhuhkqpdwpnxh43eu

AutoML Two-Sample Test [article]

Jonas M. Kübler, Vincent Stimper, Simon Buchholz, Krikamol Muandet, Bernhard Schölkopf
2022 arXiv   pre-print
We use a simple test that takes the mean discrepancy of a witness function as the test statistic and prove that minimizing a squared loss leads to a witness with optimal testing power.  ...  We provide an implementation of the AutoML two-sample test in the Python package autotst.  ...  Acknowledgments and Disclosure of Funding We thank Lisa Koch and Wittawat Jitkrittum for helpful discussions.  ... 
arXiv:2206.08843v1 fatcat:6diuvhxoxbgv5ajjcnahsaldte

Generative Models and Model Criticism via Optimized Maximum Mean Discrepancy [article]

Danica J. Sutherland, Hsiao-Yu Tung, Heiko Strathmann, Soumyajit De, Aaditya Ramdas, Alex Smola, Arthur Gretton
2021 arXiv   pre-print
We propose a method to optimize the representation and distinguishability of samples from two probability distributions, by maximizing the estimated power of a statistical test based on the maximum mean  ...  In the latter role, the optimized MMD is particularly helpful, as it gives an interpretable indication of how the model and data distributions differ, even in cases where individual model samples are not  ...  ACKNOWLEDGEMENTS We would like to thank Tim Salimans, Ian Goodfellow, and Wojciech Zaremba for providing their code and for gracious assistance in using it, as well as Jeff Schneider for helpful discussions  ... 
arXiv:1611.04488v6 fatcat:xuzejsdiw5gunjv2fzkvtkkibu

Performance analysis of comercial simulation-based optimization packages: OptQuest and Witness Optimizer

Hamidreza Eskandari, Ehsan Mahmoodi, Hamed Fallah, Christopher D. Geiger
2011 Proceedings of the 2011 Winter Simulation Conference (WSC)  
The objective of this study is to evaluate and compare two commercial simulation-based optimization packages, OptQuest and Witness Optimizer, to determine their relative performance based on the quality  ...  In Section 3, two generic test problems, the pull manufacturing system and the inventory system, are described. The experimental design is explained in Section 4.  ...  The two-sample-t test with  = 0.05 indicates that there is no difference between them.  ... 
doi:10.1109/wsc.2011.6147946 dblp:conf/wsc/EskandariMFG11 fatcat:46v6ofb6prb2hnirjozv4lvgg4

Informative Features for Model Comparison [article]

Wittawat Jitkrittum, Heishiro Kanagawa, Patsorn Sangkloy, James Hays, Bernhard Schölkopf, Arthur Gretton
2018 arXiv   pre-print
We propose two new statistical tests which are nonparametric, computationally efficient (runtime complexity is linear in the sample size), and interpretable.  ...  Given two candidate models, and a set of target observations, we address the problem of measuring the relative goodness of fit of the two models.  ...  The learning rate is set to 10 −3 (for both discriminator and generator in the two models). Some samples generated from the two trained models are shown in Figure 6 .  ... 
arXiv:1810.11630v1 fatcat:ogsh2onfcrflxms6warz2qdhju

Experimental demonstration of an efficient, semi-device-independent photonic indistinguishability witness [article]

Reinier van der Meer, Peter Hooijschuur, Franciscus H.B. Somhorst, Pim Venderbosch, Michiel de Goede, Ben Kassenberg, Henk Snijders, Caterina Taballione, Jorn Epping, Hans van den Vlekkert, Nathan Walk, Pepijn W.H. Pinkse (+1 others)
2021 arXiv   pre-print
Existing indistinguishability witnesses may be vulnerable to implementation loopholes, showing the need for a measurement which depends on as few assumptions as possible.  ...  Here, we introduce a semi-device-independent witness of photonic indistinguishability and measure it on an integrated photonic processor, certifying three-photon indistinguishability in a way that is insensitive  ...  ACKNOWLEDGEMENTS We thank Mattia Walschaers for discussions.  ... 
arXiv:2112.00067v1 fatcat:odcwy2dymrdv5hds4y73u4irca

Neural Stein critics with staged L^2-regularization [article]

Matthew Repasky, Xiuyuan Cheng, Yao Xie
2022 arXiv   pre-print
While recent studies revealed that the optimal L^2-regularized Stein critic equals the difference of the score functions of two probability distributions up to a multiplicative constant, we investigate  ...  Metrics that quantify the disparity in probability distributions, such as the Stein discrepancy, play an important role in statistical testing in high dimensions.  ...  M.R. and Y.X. are supported by an NSF CAREER Award CCF-1650913, NSF DMS-2134037, CMMI-2015787, DMS-1938106, and DMS-1830210.  ... 
arXiv:2207.03406v1 fatcat:xxszyhm6zjgwlalqqqmcugvacy

Complete nonclassicality test with a photon-number-resolving detector

T. Kiesel, W. Vogel
2012 Physical Review A. Atomic, Molecular, and Optical Physics  
We present a method for the experimental measurement of nonclassicality witnesses and demonstrate its application.  ...  This setup allows a complete test of nonclassicality of an arbitrary quantum state. The role of the quantum efficiency as well as statistical and systematic uncertainties are discussed.  ...  In any case, the test procedure only requires photon-number resolved measurements in order to estimate the witness for an arbitrary amplitude α.  ... 
doi:10.1103/physreva.86.032119 fatcat:ani63633c5gdpjqa6b35t3cygu

Efficient verification of bosonic quantum channels via benchmarking [article]

Ya-Dong Wu, Barry C. Sanders
2019 arXiv   pre-print
To this end, we construct an average-fidelity witness that yields a tight lower bound for average fidelity plus a general framework for verifying optimal quantum channels.  ...  For both multi-mode unitary Gaussian channels and single-mode amplification channels, we present experimentally feasible average-fidelity witnesses and reliable verification schemes, for which sample complexity  ...  As estimating the mean value of an average-fidelity witness is sampling the mean value of an unknown distribution, we use sampling complexities, instead of query complexities, from now on, to infer how  ... 
arXiv:1904.10682v1 fatcat:kgzrmz5bgjgfvmwk5yfbuuaeem

Using Full Field Data to Produce a Single Indentation Test for Fully Characterising the Mooney-Rivlin Material Model

John D. Van Tonder, Martin P. Venter, Gerhard Venter, S. Skatulla
2021 MATEC Web of Conferences  
This paper proposes a method of inverse finite element analysis operating under the assumption of equally objective function optimal planes or "hyper-planes".  ...  A theoretical testing method for fully characterising the Mooney-Rivlin hyper-elastic material model is proposed by capturing full-field data, namely displacement field and indentation force data.  ...  the two planes for the same indentation test at different depth the parameters are convergent.  ... 
doi:10.1051/matecconf/202134700029 fatcat:sgp6g5lyxbgvpfdjjv76gox6ky

Reduction of transient noise artifacts in gravitational-wave data using deep learning [article]

Kentaro Mogushi
2021 arXiv   pre-print
Mitigation of glitches is crucial for improving GW signal detectability.  ...  To increase the available data period and improve the detectability for both model and unmodeled GW signals, we present a new machine learning based method which uses on-site sensors/system-controls monitoring  ...  The author would like to thank their LIGO Scientific Collaboration and Virgo Collaboration colleagues for their help and useful comments, in particular Yanyan  ... 
arXiv:2105.10522v2 fatcat:mgzfpbpczjcv7gy6tsknp5xgcq

Classification Logit Two-sample Testing by Neural Networks [article]

Xiuyuan Cheng, Alexander Cloninger
2020 arXiv   pre-print
This paper proposes a two-sample statistic which is the difference of the logit function, provided by a trained classification neural network, evaluated on the testing set split of the two datasets.  ...  Network-based tests have the computational advantage that the algorithm scales to large samples.  ...  We call f * the population witness function of the logit test. The witness function plays an important role in the ability of the test to distinguish two densities.  ... 
arXiv:1909.11298v2 fatcat:2slyeictivhuvghu2jnaexccze

RRP Nb$_{3}$Sn Strand Studies for LARP

Emanuela Barzi, Rodger Bossert, Shlomo Caspi, Daniel R. Dietderich, Paolo Ferracin, Arup Ghosh, Daniele Turrioni
2007 IEEE transactions on applied superconductivity  
Using strand billet qualification and tests of strands extracted from cables, the short sample limits of magnet performance were obtained.  ...  The Nb 3 Sn strand chosen for the next step in the magnet R&D of the U.S.  ...  ACKNOWLEDGMENT The authors thank Lance Cooley for his contribution to the heat treatment studies.  ... 
doi:10.1109/tasc.2007.899579 fatcat:gtl2ec2f6jebpnks2jz54az3ya

Learning to Superoptimize Real-world Programs [article]

Alex Shypula, Pengcheng Yin, Jeremy Lacomis, Claire Le Goues, Edward Schwartz, Graham Neubig
2022 arXiv   pre-print
We created a dataset consisting of over 25K real-world x86-64 assembly functions mined from open-source projects and propose an approach, Self Imitation Learning for Optimization (SILO) that is easy to  ...  Our method, SILO, superoptimizes 5.9% of our test set when compared with the gcc version 10.3 compiler's aggressive optimization level -O3.  ...  For each input specification S i in B ex , we sample a model-predicted optimization Fi o , and execute Fi o on the I/O test suite {IO} K k=1 to compute the cost function C.  ... 
arXiv:2109.13498v2 fatcat:fp4j7bhiyja2vf7tctpcfj4u6y

Learning Proximal Operators to Discover Multiple Optima [article]

Lingxiao Li, Noam Aigerman, Vladimir G. Kim, Jiajin Li, Kristjan Greenewald, Mikhail Yurochkin, Justin Solomon
2022 arXiv   pre-print
We present an end-to-end method to learn the proximal operator across a family of non-convex problems, which can then be used to recover multiple solutions for unseen problems at test time.  ...  We further present a benchmark for multi-solution optimization including a wide range of applications and evaluate our method to demonstrate its effectiveness.  ...  We sample 1024 witnesses to compute WP δ t , averaged over 256 test problem instances.  ... 
arXiv:2201.11945v1 fatcat:fzet2z2ye5d6za5yimbrtjr5cy
« Previous Showing results 1 — 15 out of 122,731 results