307 Hits in 6.2 sec

Asymptotic statistical theory of overtraining and cross-validation

S. Amari, N. Murata, K.-R. Muller, M. Finke, H.H. Yang
1997 IEEE Transactions on Neural Networks  
A statistical theory for overtraining is proposed.  ...  Although cross-validated early stopping is useless in the asymptotic region, it surely decreases the generalization error in the nonasymptotic region.  ...  Schulten for warm hospitality during his stay at the Beckman Inst. in Urbana, IL. and for warm hospitality at RIKEN during the completion of this work.  ... 
doi:10.1109/72.623200 pmid:18255701 fatcat:qjkqoolfbrcazlvkxc2iddauey

Page 1105 of Neural Computation Vol. 8, Issue 5 [page]

1996 Neural Computation  
Statistical theory of overtraining—lIs cross-validation effective? In Advances in Neural Information Processing Systems 8, D. S. Touretzky, M. C. Mozer, and M. E. Has- selmo, eds.  ...  Asymptotic statistical theory of overtraining and cross-validation. University of Tokyo Tech. Rep. METR 95-06. IEEE Trans. Neural Networks (submitted). Amari, S., Murata, N., Miiller, K.  ... 

Maximum Entropy and Learning Theory

Griff L. Bilbro, David E. Van den Bout
1992 Neural Computation  
It is difficult to obtain large overtraining effects from this theory even though some traces of overtraining remain: The curves of Figure 3 can be made to cross for small m and poor prior.  ...  This is important because in actual training, the effect of the initial configuration is also quickly lost. For m < 20 the predictions are not valid in any case, since 852 Griff L.  ... 
doi:10.1162/neco.1992.4.6.839 fatcat:wjuwcirijfc65lkbafppu3ncmu

Overshadowing as a function of trial number: Dynamics of first- and second-order comparator effects

Steven Stout, Francisco Arcediano, Martha Escobar, Ralph R. Miller
2003 Animal Learning and Behavior  
In two conditioned lick suppression experiments with rats, we examined the permanence of the overshadowing effect as a function of the number of compound reinforced training trials.  ...  Experiment 2 extended the generality of the effect to a sensory preconditioning design and further demonstrated that overshadowing lost through many compound-US pairings was restored by posttraining extinction  ...  Stimulus preexposure was included to equate the subject's experience with that of the subjects in Experiment 1, thereby facilitating the external validity of possible cross-experiment comparisons.  ... 
doi:10.3758/bf03195972 fatcat:vybpq46mkvcrzdui5y644tvq3i

A Numerical Study on Learning Curves in Stochastic Multilayer Feedforward Networks

K.-R. Müller, M. Finke, N. Murata, K. Schulten, S. Amari
1996 Neural Computation  
Statistical theory of overtraining—lIs cross-validation effective? In Advances in Neural Information Processing Systems 8, D. S. Touretzky, M. C. Mozer, and M. E. Has- selmo, eds.  ...  Asymptotic statistical theory of overtraining and cross-validation. University of Tokyo Tech. Rep. METR 95-06. IEEE Trans. Neural Networks (submitted). Amari, S., Murata, N., Miiller, K.  ... 
doi:10.1162/neco.1996.8.5.1085 pmid:8697228 fatcat:y7ildqvb7javdco5vrq7dr4lkm

Particle Swarm Optimization of Fuzzy ARTMAP Parameters

E. Granger, P. Henniges, L.S. Oliveira, R. Sabourin
2006 The 2006 IEEE International Joint Conference on Neural Network Proceedings  
Furthermore, the PSO strategy eliminates degradation of generalization error due to overtraining resulting from the training set size, number of training epochs, and data set structure.  ...  In this paper a Particle Swarm Optimization (PSO)-based training strategy is introduced for fuzzy ARTMAP that minimizes generalization error while optimizing parameter values.  ...  These are classic parametric and non-parametric classification techniques from statistical pattern recognition, which are immune to the effects of overtraining.  ... 
doi:10.1109/ijcnn.2006.246975 dblp:conf/ijcnn/GrangerHOS06 fatcat:ob32oregungqtf4xzhvn2xljou

The Effects of Adding Noise During Backpropagation Training on a Generalization Performance

Guozhong An
1996 Neural Computation  
The shortcoming of the validation-set approach is that it is effective only if both the train- ing set and the validation set are large and representative.  ...  This finding is also in line with the theory that penalizing large derivatives at the sigmoidal units weakens the effect of penalizing large hidden-layer activation.  ... 
doi:10.1162/neco.1996.8.3.643 fatcat:kg32434xo5apvb4pq2zkhowfiq

Learning-Induced Enduring Changes in Functional Connectivity among Prefrontal Cortical Neurons

E. H. Baeg, Y. B. Kim, J. Kim, J.-W. Ghim, J. J. Kim, M. W. Jung
2007 Journal of Neuroscience  
when learning reached asymptote.  ...  The present finding of negatively accelerated changes in FC confirms associative learning theories and provides crucial neurophysiological evidence for Hebb's hypothesis.  ...  In the overtraining phase, animals performed both tasks at asymptotic levels ( Fig. 1G ).  ... 
doi:10.1523/jneurosci.4759-06.2007 pmid:17251433 fatcat:vsxy27ymwvhizmicj7zhsv4wfi

Optimized approach to decision fusion of heterogeneous data for breast cancer diagnosis

Jonathan L. Jesneck, Loren W. Nolte, Jay A. Baker, Carey E. Floyd, Joseph Y. Lo
2006 Medical Physics (Lancaster)  
This dissertation presents a computer aid known as optimized decision fusion, and explores both its underlying theory and practical application.  ...  , and probit), Bayesian model averaging of these regression models, least angle regression, and a support vector machine.  ...  We explored likelihood ratio theory and its effects on decision fusion.  ... 
doi:10.1118/1.2208934 pmid:16964873 pmcid:PMC2569003 fatcat:av3yxepgordtjhdwn42ssv2ti4

Parametric Statistical Estimation with Artificial Neural Networks: A Condensed Discussion [chapter]

Halbert White
1994 From Statistics to Neural Networks  
Learning in artificial neural networks is a process by which experience arising from exposure to measurements of empirical phenomena is converted to knowledge, embodied in network weights.  ...  We exploit this formal viewpoint to give a unified theory of learning in artificial neural networks.  ...  Acknowledgements Support from IBM Corporation and the National Science Foundation under grants SES-8921382 and SES-9209023 is gratefully acknowledged.  ... 
doi:10.1007/978-3-642-79119-2_6 fatcat:yt4tgtenrrb3xkrr5efwgvuwzu

A comparative evaluation of the generalised predictive ability of eight machine learning algorithms across ten clinical metabolomics data sets for binary classification

Kevin M. Mendez, Stacey N. Reinke, David I. Broadhurst
2019 Metabolomics  
Metabolomics is increasingly being used in the clinical setting for disease diagnosis, prognosis and risk prediction.  ...  Machine learning algorithms are particularly important in the construction of multivariate metabolite prediction.  ...  A full parameter search was performed for each individual model under cross-validation conditions, and repeatedly the same hyperparameters had little effect on optimisation, so for clarity of presentation  ... 
doi:10.1007/s11306-019-1612-4 pmid:31728648 pmcid:PMC6856029 fatcat:7uze5moxgraspgpueu6sfpuuqa

The paradox of hematocrit in exercise physiology: which is the "normal" range from an hemorheologist's viewpoint?

J F Brun, C Bouchahda, D Chaze, A A Benhaddad, J P Micallef, J Mercier
2000 Clinical Hemorheology and Microcirculation  
The paradox of hematocrit in exercise physiology is that artificially increasing it by autotransfusion or erythropoietin doping improves VO2 max and performance, while in normal conditions there is a strong  ...  We aimed at investigating: (a) which is the physiological range of hematocrit in highly trained professional footballers; (b) what are the characteristics of athletes with high vs low hematocrit?  ...  No cross reactivity with IGFBP-2, 3 and 4 has been detected. Statistics Results are presented as mean ± the SE of the mean. Normality of data distribution was tested by the Kolmogorov-Smirnov test.  ... 
pmid:11081466 fatcat:nvcsw5vq6nfrrixbr2f6mirbce

Anomaly Detection for Resonant New Physics with Machine Learning

Jack Collins, Kiel Howe, Benjamin Nachman
2018 Physical Review Letters  
Therefore, it is essential to broaden the sensitivity of the search program to include unexpected scenarios.  ...  The only requirement on the signal for this new procedure is that it is localized in at least one known direction in phase space.  ...  The effects of overtraining on the performance of the classifiers can be mitigated by a nested crossvalidation procedure, as is described in detail in Ref. [37] .  ... 
doi:10.1103/physrevlett.121.241803 fatcat:yialkdnvqjdo5c7cthu3wuiwty

Classifier's Complexity Control while Training Multilayer Perceptrons [chapter]

Šarūnas Raudys
2000 Lecture Notes in Computer Science  
A special attention is paid to magnitudes of the weights and to optimization of the training procedure.  ...  Here qualities of statistical and neural net approaches are merged together.  ...  Typically in situations with unknown data, the variance is determined by the cross validation technique.  ... 
doi:10.1007/3-540-44522-6_4 fatcat:zo6dbjj5lfgrlkjrdzv23oydfq

Rethinking generalization requires revisiting old ideas: statistical mechanics approaches and complex learning behavior [article]

Charles H. Martin, Michael W. Mahoney
2019 arXiv   pre-print
Using this model, we describe how a very simple application of ideas from the statistical mechanics theory of generalization provides a strong qualitative description of recently-observed empirical results  ...  Within this approach, we present a prototypical Very Simple Deep Learning (VSDL) model, whose behavior is controlled by two control parameters, one describing an effective amount of data, or load, on the  ...  See [9, 30, 31] for more comprehensive introductions; and see also [10] for an interesting discussion of the use of SM for cross validation.)  ... 
arXiv:1710.09553v2 fatcat:h2royn4g4rcexhlxugqgrh7pqu
« Previous Showing results 1 — 15 out of 307 results