3,623 Hits in 9.3 sec

Pruning Neural Networks at Initialization: Why are We Missing the Mark? [article]

Jonathan Frankle, Gintare Karolina Dziugaite, Daniel M. Roy, Michael Carbin
2021 arXiv   pre-print
Recent work has explored the possibility of pruning neural networks at initialization.  ...  This property suggests broader challenges with the underlying pruning heuristics, the desire to prune at initialization, or both.  ...  We do not identify a cause for why these methods struggle to prune in a specific fashion at initialization (and often only at initialization), and we believe that this is an important question for future  ... 
arXiv:2009.08576v2 fatcat:ki5se6xa3ranzpc5cbj3o5iy3i

MARVEL - D3.1: Multimodal and privacy-aware audio-visual intelligence – initial version

Alexandros Iosifidis
2022 Zenodo  
This document describes the initial version of the methodologies pro- posed by MARVEL partners towards the realisation of the Audio, Visual and Multimodal AI Subsystem of the MARVEL architecture.  ...  The effectiveness of these methods is compared against recent baselines, towards achieving the AI methodology- related objectives of the MARVEL project.  ...  Performance Evaluation We benchmarked the two proposed architectures, for waveforms and spectrograms, on the UrbanSound8K dataset [22].  ... 
doi:10.5281/zenodo.6821317 fatcat:eia7rkk5lfbg7khs3qcat5qd3m

Autonomous mobile robot global self-localization using Kohonen and region-feature neural networks

Jason A. Jan�t, Ricardo Gutierrez, Troy A. Chase, Mark W. White, John C. Sutton
1997 Journal of Robotic Systems  
This article presents and compares two neural network-based approaches to global selflocalization (GSL) for autonomous mobile robots using: (1) a Kohonen neural network; and (2) a region-feature neural  ...  The classification rate of both approaches are comparable and, thus, worthy of presentation. The observed pros and cons of both approaches are also discussed.  ...  Figure 9 marks with an ϫ all nodes that are pruned.  ... 
doi:10.1002/(sici)1097-4563(199704)14:4<263::aid-rob4>;2-o fatcat:ho5zrih3p5glbgbmycvhxuq7fi

Regularized extreme learning machine for regression with missing data

Qi Yu, Yoan Miche, Emil Eirola, Mark van Heeswijk, Eric Séverin, Amaury Lendasse
2013 Neurocomputing  
performance by the two penalties, Moreover, it completes ELM with a new tool to solve missing data problem even when half of the training data are missing as the extreme case.  ...  This paper proposes a method which is the advanced modification of the original Extreme Learning Machine with a new tool to solve the missing data problem.  ...  If data are not missing at random or completely at random then they are classed as Missing Not at Random (MNAR). When we have data that are MNAR we have a problem.  ... 
doi:10.1016/j.neucom.2012.02.040 fatcat:csxagwea5bb3lkxidzbp5pzf6m

DAEMON Deliverable 2.2: Initial DAEMON Network Intelligence framework and toolsets

Antonio Bazco Nogueras, Marco Fiore, Ginés García Avilés, Michele Gucciardo, Miguel Camelo, Paola Soto, Chia-Yu Chang, Danny De Vleeschauwer, Josep Xavier Salvat Lozano, Andrés García Saavedra, Xi Li, Alexandros Kostopoulos (+7 others)
2022 Zenodo  
This is a live effort that aims at keeping the project activities up to date with the current state of the art in scien [...]  ...  These updates are mainly driven by the feedback provided by the implementation and evaluation activities of the project to date, which unveiled the need for adjustments to the original requirements or  ...  NIP requirement tree are marked as relevant to all KPIs targeted by the project.  ... 
doi:10.5281/zenodo.6970839 fatcat:h43prrnjvzadxhg72iulko7bkq

Affective Monitoring: A Generic Mechanism for Affect Elicitation

R. Hans Phaf, Mark Rotteveel
2012 Frontiers in Psychology  
Affective monitoring is considered a local match-mismatch process within a module of the neural network.  ...  For affect to become conscious, large-scale synchronization of the oscillations over the network and the construction of emotional experiences are required.  ...  ACKNOWLEDGMENTS We are grateful to Bram T. Heerebout, A. E. Yoram Tap, and William H. Thompson for their help in various stages of this work.  ... 
doi:10.3389/fpsyg.2012.00047 pmid:22403557 pmcid:PMC3290827 fatcat:6h3ngbz3v5f5pnq5pxvysf56w4

Adolescent impulsivity phenotypes characterized by distinct brain networks

Robert Whelan, Patricia J Conrod, Jean-Baptiste Poline, Anbarasu Lourdusamy, Tobias Banaschewski, Gareth J Barker, Mark A Bellgrove, Christian Büchel, Mark Byrne, Tarrant D R Cummins, Mira Fauth-Bühler, Herta Flor (+20 others)
2012 Nature Neuroscience  
Hypofunctioning of a specific orbitofrontal cortical network was associated with likelihood of initiating drug use in early adolescence.  ...  As these networks have not yet been described, we identified distinct cortical and subcortical networks underlying successful inhibitions and inhibition failures in a large sample (n = 1,896) of 14-year-old  ...  Notably, these findings also suggest a degree of independence at the neural level for the deficits in inhibitory control associated with ADHD and the propensity for drug abuse.  ... 
doi:10.1038/nn.3092 pmid:22544311 fatcat:a6cfc7cywvadtco4xxz6vmkmz4

Spikelets in Pyramidal Neurons: Action Potentials Initiated in the Axon Initial Segment That Do Not Activate the Soma

Martina Michalikova, Michiel W. H. Remme, Richard Kempter, Volker Steuber
2017 PLoS Computational Biology  
Therefore, APs are the basis of neural function, yet some of their fundamental features are still not well understood.  ...  Here, we propose that spikelets occur when an AP initiated at the axon initial segment only propagates down the axon, but fails to activate sodium currents in the soma and dendrites.  ...  Acknowledgments We thank M. Brecht, E. Chorev, N. Maier, S. Schreiber, and R. D. Traub for valuable discussions.  ... 
doi:10.1371/journal.pcbi.1005237 pmid:28068338 pmcid:PMC5221759 fatcat:f2suho3umzbv5hucj6fannkz4e

Language and connectionism: the developing interface

Mark S. Seidenberg
1994 Cognition  
After a difficult initial period in which connectionism was perceived as either irrelevant or antithetical to linguistic theory, connectionist concepts are now beginning to be brought to bear on basic  ...  I consider how connectionist concepts may be relevant to issues concerning the representation of linguistic knowledge; the role of a priori constraints on acquisition; and the poverty of the stimulus argument  ...  This level of description is useful but misses certain generalizations that can only be stated at the level of network behavior.  ... 
doi:10.1016/0010-0277(94)90037-x pmid:8039370 fatcat:pn2ddhwvzfasplafbwkyl5xxu4

Simon Says: Evaluating and Mitigating Bias in Pruned Neural Networks with Knowledge Distillation [article]

Cody Blakeney, Nathaniel Huish, Yan Yan, Ziliang Zong
2021 arXiv   pre-print
Third, we reveal that model similarity has strong correlations with pruning induced bias, which provides a powerful method to explain why bias occurs in pruned neural networks.  ...  A clear gap exists in the current literature on evaluating and mitigating bias in pruned neural networks.  ...  Moreover, we reveal that model similarity has strong correlations with pruning induced bias, which can be used to explain why bias occurs in pruned neural networks.  ... 
arXiv:2106.07849v1 fatcat:gsj4lnxicrgvjfzomxau3bw24m

Demystifying Neural Network Filter Pruning [article]

Zhuwei Qin, Fuxun Yu, ChenChen Liu, Xiang Chen
2018 arXiv   pre-print
L1 norm), conventional filter pruning methods for Convolutional Neural Networks (CNNs) have been proved with great effectiveness in computation load reduction.  ...  In this work, we explore the filter pruning and the retraining through qualitative filter functionality interpretation.  ...  As shown in the figure, the filters' visualized patterns are ranked by the 1 norm in an ascending order, where the pruned filters are marked by red slashes.  ... 
arXiv:1811.02639v1 fatcat:wpublreeijh4pix2m5no5btmoi

What it means to be Zen: Marked modulations of local and interareal synchronization during open monitoring meditation

Anne Hauswald, Teresa Übelacker, Sabine Leske, Nathan Weisz
2015 NeuroImage  
We found a correlation between graph measures in the 160-170 Hz range and MAAS scores.  ...  In the present study, we took advantage of this ability and studied brain activity related to the shift of mental state. Electrophysiological activity, i.e.  ...  Acknowledgments The work was supported by the Deutsche Forschungsgemeinschaft Grant WE 4156/2-1. The contribution by NW was supported by the European Research Council Grant WIN2CON, ERC StG 283404.  ... 
doi:10.1016/j.neuroimage.2014.12.065 pmid:25562827 pmcid:PMC4359616 fatcat:n32kziqp3baytcmaj7mq3uofne

The structure dilemma in biological and artificial neural networks

Thomas Pircher, Bianca Pircher, Eberhard Schlücker, Andreas Feigenspan
2021 Scientific Reports  
We trained these diverse networks, which markedly differ in their architecture, initialization and pruning technique, and we found remarkable parallels between biological-based and artificial neural networks  ...  , as we were additionally able to show that the dilemma is also present in artificial neural networks.  ...  Acknowledgements We thank Dominik Haspel (Institute of Process Machinery and Systems Engineering, Friedrich-Alexander University Erlangen-Nuremberg) for his inspiring and helpful comments and technical  ... 
doi:10.1038/s41598-021-84813-6 pmid:33692408 pmcid:PMC7970964 fatcat:5wmol5gnv5g43jjhsiqk3yops4

Fuzzification and Reduction of Information-Theoretic Rule Sets [chapter]

Mark Last, Abraham Kandel
2001 Studies in Fuzziness and Soft Computing  
We start with constructing an information-theoretic network from a data table and extracting a set of association rules based on the network connections.  ...  If-then rules are one of the most common forms of knowledge discovered by data mining methods.  ...  Acknowledgment This work was partially supported by the USF Center for Software Testing under grant no. 2108-004-00.  ... 
doi:10.1007/978-3-7908-1825-3_3 fatcat:lwgjiwa3cbhifkvtwajjrjqkni

Dynamic systems modelling using genetic programming

Mark P. Hinchliffe, Mark J. Willis
2003 Computers and Chemical Engineering  
A comparison with neural networks reveals that GP is able to match the accuracy of the network predictions but is more expensive computationally.  ...  Two case studies are used to demonstrate the performance of this technique. The first example uses the multi-objective algorithm to improve the parsimony of the evolved model structures.  ...  initial neural network prediction error is substantially higher than that of the GP algorithm.  ... 
doi:10.1016/j.compchemeng.2003.06.001 fatcat:jooelmnx75dmtppputp2vky7hq
« Previous Showing results 1 — 15 out of 3,623 results