Filters








153,855 Hits in 9.5 sec

One Method to Rule Them All: Variance Reduction for Data, Parameters and Many New Methods [article]

Filip Hanzely, Peter Richtárik
2020 arXiv   pre-print
In special cases, our method reduces to several known and previously thought to be unrelated methods, such as SAGA, LSVRG, JacSketch, SEGA and ISEGA, and their arbitrary sampling and proximal generalizations  ...  As a by-product, we provide the first unified method and theory for stochastic gradient and stochastic coordinate descent type methods.  ...  Appendix One Method to Rule Them All: Variance Reduction for Data, Parameters and Many New Methods A Table of Contents For easier navigation through the paper and the appendices, we include a table  ... 
arXiv:1905.11266v2 fatcat:hnhbmvpsljf3rkt43g7j3gu4dy

One algorithm to rule them all? An evaluation and discussion of ten eye movement event-detection algorithms

Richard Andersson, Linnea Larsson, Kenneth Holmqvist, Martin Stridh, Marcus Nyström
2016 Behavior Research Methods  
Differing results across evaluation methods make it difficult to select one winner for fixation detection.  ...  Almost all eye-movement researchers use algorithms to parse raw data and detect distinct types of eye movement events, such as fixations, saccades, and pursuit, and then base their results on these.  ...  According to Salvucci and Goldberg (2000) , it is based on the data reduction algorithm by Widdel (1984) .  ... 
doi:10.3758/s13428-016-0738-9 pmid:27193160 fatcat:tnc7fiq7ufa3tnpggk5ogosc4y

Variance Reduction in Deep Learning: More Momentum is All You Need [article]

Lionel Tondji, Sergii Kashubin, Moustapha Cisse
2021 arXiv   pre-print
Our proposal leads to faster convergence than vanilla methods on standard benchmark datasets (e.g., CIFAR and ImageNet). It is robust to label noise and amenable to distributed optimization.  ...  Variance reduction (VR) techniques have contributed significantly to accelerating learning with massive datasets in the smooth and strongly convex setting (Schmidt et al., 2017; Johnson & Zhang, 2013;  ...  g (n) t + N k=1 p k g (k) t (4) The update rule stems from the traditional use of control variates for variance reduction (Fishman, 1996) with the additional trick to use one control variate for each  ... 
arXiv:2111.11828v1 fatcat:er5drxbzercvbn3q4xgaud4ykm

Reinforcement Learning When All Actions Are Not Always Available

Yash Chandak, Georgios Theocharous, Blossom Metevier, Philip Thomas
2020 PROCEEDINGS OF THE THIRTIETH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE AND THE TWENTY-EIGHTH INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE CONFERENCE  
In this paper we argue that existing RL algorithms for SAS-MDPs can suffer from potential divergence issues, and present new policy gradient algorithms for SAS-MDPs that incorporate variance reduction  ...  techniques unique to this setting, and provide conditions for their convergence.  ...  We are also immensely grateful to the three anonymous reviewers who shared their insights and feedback, specially to the second reviewer who helped improve the counter example.  ... 
doi:10.1609/aaai.v34i04.5740 fatcat:v23t5iiwdzdrppwqcd7yto4ca4

Reinforcement Learning When All Actions are Not Always Available [article]

Yash Chandak, Georgios Theocharous, Blossom Metevier, Philip S. Thomas
2020 arXiv   pre-print
In this paper we argue that existing RL algorithms for SAS-MDPs can suffer from potential divergence issues, and present new policy gradient algorithms for SAS-MDPs that incorporate variance reduction  ...  techniques unique to this setting, and provide conditions for their convergence.  ...  We are also immensely grateful to the three anonymous reviewers who shared their insights and feedback, specially to the second reviewer who helped improve the counter example.  ... 
arXiv:1906.01772v2 fatcat:tp6gzx2k7bdtdkxryo6jzpct5y

All-Action Policy Gradient Methods: A Numerical Integration Approach [article]

Benjamin Petit, Loren Amdahl-Culleton, Yao Liu, Jimmy Smith, Pierre-Luc Bacon
2019 arXiv   pre-print
In this paper, we adopt a numerical integration perspective to broaden the applicability of the all-action estimator to general spaces and to any function class for the policy or critic components, beyond  ...  When this integral can be computed, the resulting "all-action" estimator [Sutton, 2001] provides a conditioning effect [Bratley, 1987] reducing the variance significantly compared to the REINFORCE estimator  ...  Our methods could be extended in many ways, notably by combining them with other variance reduction techniques such as control variates.  ... 
arXiv:1910.09093v1 fatcat:gz6xsivpivhk3knjmxnl6zmbqa

Enumerating all maximal biclusters in numerical datasets [article]

Rosana Veroneze, Arindam Banerjee, Fernando J. Von Zuben
2015 arXiv   pre-print
Biclustering has proved to be a powerful data analysis technique due to its wide success in various application domains.  ...  Here, we present a general family of biclustering algorithms for enumerating all maximal biclusters with (i) constant values on rows, (ii) constant values on columns, or (iii) coherent values.  ...  For each one of them, we remove the empty spots; we threw away the data for any genes where one or more expression levels were not measured; we filtered out genes with small variance over time; and  ... 
arXiv:1403.3562v4 fatcat:t5eogz6xv5a5hjl6fbbvpiiaxa

Enumeration of Time Series Motifs of All Lengths

Abdullah Mueen
2013 2013 IEEE 13th International Conference on Data Mining  
For all such purposes, many high quality motifs of various lengths are desirable and thus, originates the problem of enumerating motifs for a wide range of lengths.  ...  The algorithm frees us from re-discovering the same motif at different lengths and tuning multiple data-dependent parameters.  ...  None of them guarantees finding motifs of various lengths and all of them require a set of data-dependent parameters.  ... 
doi:10.1109/icdm.2013.27 dblp:conf/icdm/Mueen13 fatcat:na5mmefwjvfktpmwibhb5m2e3i

Unsupervised Feature Learning With Winner-Takes-All Based STDP

Paul Ferré, Franck Mamalet, Simon J. Thorpe
2018 Frontiers in Computational Neuroscience  
We apply our method to extract features from the MNIST, ETH80, CIFAR-10, and STL-10 datasets and show that these features are relevant for classification.  ...  We show equivalence between rank order coding Leaky-Integrate-and-Fire neurons and ReLU artificial neurons when applied to non-temporal data.  ...  ACKNOWLEDGMENTS We would like to thank Timothée Masquelier, Saeed Reza Kheradpisheh, Douglas McLelland, Christophe Garcia, and Stefan Dufner for their advice on the method and the manuscript.  ... 
doi:10.3389/fncom.2018.00024 pmid:29674961 pmcid:PMC5895733 fatcat:hxdcric4abg5jiqims53p23pwi

Skin Cancer Diagnostics with an All-Inclusive Smartphone Application

Kalwa, Legner, Kong, Pandey
2019 Symmetry  
This all-inclusive smartphone application is designed to be easy-to-download and easy-to-navigate for the end user, which is imperative for the eventual democratization of such medical diagnostic systems  ...  By using adaptive algorithms in the individual data-processing stages, our approach is made computationally light, user friendly, and reliable in discriminating melanoma cases from benign ones.  ...  Acknowledgments: The authors are grateful to the image database (PH2 and MED-NODE) for providing public access to the images. This work is partially supported by the U.S.  ... 
doi:10.3390/sym11060790 fatcat:4hmn4rze2zbg3b34eu5vglqxi4

Differential Privacy: What is all the noise about? [article]

Roxana Danger
2022 arXiv   pre-print
DP has been actively researched during the last 15 years, but it is still hard to master for many Machine Learning (ML)) practitioners.  ...  This paper aims to provide an overview of the most important ideas, concepts and uses of DP in ML, with special focus on its intersection with Federated Learning (FL).  ...  Many other defence methods are available in the literature (see [61] for a comprehensive list of them), but unlike DP, SNPC and HE, they are tailored to specific types of attacks and therefore less useful  ... 
arXiv:2205.09453v1 fatcat:5z3nqsh7qbbwfhbrc6hmzt43ya

Behavioural choice emerges from nonlinear all-to-all interactions between drives [article]

Stephen C Thornquist, Michael A Crickmore
2020 bioRxiv   pre-print
Extending these findings to model the interactions between all of an animal's motivations led to the surprising prediction that, under many conditions, all-to-all interactions actually buffer the dominant  ...  We experimentally validated this prediction, showing that weak drives for a variety of tertiary goals can have a profound stabilizing effect on the ongoing behaviour.  ...  is the strength of the demotivating input) to estimate the parameters of the model 171 (Extended Data Figure 3a , see Methods for more information).  ... 
doi:10.1101/2020.03.12.989574 fatcat:xee73kdq7rgcdl3pvl5mlm2xdy

Modern Applied Science, Vol. 3, No. 8, August 2009, all in one file

MAS Editor
2009 Modern Applied Science  
Acknowledgements A special thank to FRIM for securing grant for this project and to members of FRIM and Faculty of Forestry, Universiti Putra Malaysia, for their supervision in this research work.  ...  Further study on improving the mechanical and physical performances of the treated particleboards should be conducted to expand the usage of the particleboards.  ...  Complete loss of data, and delete the duplicate object, then structure the decision Third. Extract rules. Extract valuable rules according to the minimum reduction getting ahead.  ... 
doi:10.5539/mas.v3n8p0 fatcat:qhc7i4dl6rfpjcywtasdohn2we

One Instrument to Rule Them All: The Bias and Coverage of Just-ID IV [article]

Joshua Angrist, Michal Kolesár
2022 arXiv   pre-print
Three widely-cited applications are used to explain why endogeneity is likely low enough for IV estimates to be reliable.  ...  Confidence interval undercoverage exceeds 5% only for endogeneity beyond that seen even when IV and OLS estimates differ by an order of magnitude.  ...  Our numerical calculations indicate that this fails to hold for all parameter values conditional on the estimated first stage sign.  ... 
arXiv:2110.10556v4 fatcat:x632g6sxd5bbjbthzlknn26dcq

All correlations must die: Assessing the significance of a stochastic gravitational-wave background in pulsar timing arrays

S. R. Taylor, L. Lentati, S. Babak, P. Brem, J. R. Gair, A. Sesana, A. Vecchio
2017 Physical Review D  
These methods are immediately applicable to all current pulsar-timing array datasets, and should become standard tools for future analyses.  ...  We then explore a method to null correlations between pulsars by using a "scrambled" overlap-reduction function in the signal model for the array.  ...  on the data with new phase-shifted Fourier design matrices, F , in the model for low-frequency processes.  ... 
doi:10.1103/physrevd.95.042002 fatcat:ilaopniginccxoqtslholxpedq
« Previous Showing results 1 — 15 out of 153,855 results