IEEE Trans Neural Netw Learn Syst - Is extreme learning machine feasible? A theoretical assessment (part II).

Tópicos

{ problem(2511) optim(1539) algorithm(950) }
{ network(2748) neural(1063) input(814) }
{ method(2212) result(1239) propos(1039) }
{ imag(1947) propos(1133) code(1026) }
{ learn(2355) train(1041) set(1003) }
{ result(1111) use(1088) new(759) }
{ model(3404) distribut(989) bayesian(671) }
{ import(1318) role(1303) understand(862) }
{ group(2977) signific(1463) compar(1072) }
{ use(2086) technolog(871) perceiv(783) }
{ data(1737) use(1416) pattern(1282) }
{ control(1307) perform(991) simul(935) }
{ measur(2081) correl(1212) valu(896) }
{ featur(3375) classif(2383) classifi(1994) }
{ surgeri(1148) surgic(1085) robot(1054) }
{ risk(3053) factor(974) diseas(938) }
{ first(2504) two(1366) second(1323) }
{ can(981) present(881) function(850) }
{ method(984) reconstruct(947) comput(926) }
{ search(2224) databas(1162) retriev(909) }
{ age(1611) year(1155) adult(843) }
{ cost(1906) reduc(1198) effect(832) }
{ sampl(1606) size(1419) use(1276) }
{ high(1669) rate(1365) level(1280) }
{ method(1219) similar(1157) match(930) }
{ chang(1828) time(1643) increas(1301) }
{ clinic(1479) use(1117) guidelin(835) }
{ data(1714) softwar(1251) tool(1186) }
{ general(901) number(790) one(736) }
{ howev(809) still(633) remain(590) }
{ data(3963) clinic(1234) research(1004) }
{ model(2341) predict(2261) use(1141) }
{ studi(1119) effect(1106) posit(819) }
{ monitor(1329) mobil(1314) devic(1160) }
{ intervent(3218) particip(2042) group(1664) }
{ activ(1138) subject(705) human(624) }
{ patient(1821) servic(1111) care(1106) }
{ health(1844) social(1437) communiti(874) }
{ use(1733) differ(960) four(931) }
{ can(774) often(719) complex(702) }
{ inform(2794) health(2639) internet(1427) }
{ system(1976) rule(880) can(841) }
{ imag(1057) registr(996) error(939) }
{ bind(1733) structur(1185) ligand(1036) }
{ sequenc(1873) structur(1644) protein(1328) }
{ imag(2830) propos(1344) filter(1198) }
{ imag(2675) segment(2577) method(1081) }
{ patient(2315) diseas(1263) diabet(1191) }
{ take(945) account(800) differ(722) }
{ studi(2440) review(1878) systemat(933) }
{ motion(1329) object(1292) video(1091) }
{ assess(1506) score(1403) qualiti(1306) }
{ treatment(1704) effect(941) patient(846) }
{ framework(1458) process(801) describ(734) }
{ error(1145) method(1030) estim(1020) }
{ concept(1167) ontolog(924) domain(897) }
{ algorithm(1844) comput(1787) effici(935) }
{ extract(1171) text(1153) clinic(932) }
{ method(1557) propos(1049) approach(1037) }
{ design(1359) user(1324) use(1319) }
{ model(2220) cell(1177) simul(1124) }
{ care(1570) inform(1187) nurs(1089) }
{ featur(1941) imag(1645) propos(1176) }
{ case(1353) use(1143) diagnosi(1136) }
{ studi(1410) differ(1259) use(1210) }
{ perform(999) metric(946) measur(919) }
{ research(1085) discuss(1038) issu(1018) }
{ system(1050) medic(1026) inform(1018) }
{ visual(1396) interact(850) tool(830) }
{ compound(1573) activ(1297) structur(1058) }
{ perform(1367) use(1326) method(1137) }
{ blood(1257) pressur(1144) flow(957) }
{ spatial(1525) area(1432) region(1030) }
{ record(1888) medic(1808) patient(1693) }
{ health(3367) inform(1360) care(1135) }
{ model(3480) simul(1196) paramet(876) }
{ ehr(2073) health(1662) electron(1139) }
{ state(1844) use(1261) util(961) }
{ research(1218) medic(880) student(794) }
{ patient(2837) hospit(1953) medic(668) }
{ model(2656) set(1616) predict(1553) }
{ data(2317) use(1299) case(1017) }
{ medic(1828) order(1363) alert(1069) }
{ signal(2180) analysi(812) frequenc(800) }
{ gene(2352) biolog(1181) express(1162) }
{ data(3008) multipl(1320) sourc(1022) }
{ time(1939) patient(1703) rate(768) }
{ analysi(2126) use(1163) compon(1037) }
{ structur(1116) can(940) graph(676) }
{ cancer(2502) breast(956) screen(824) }
{ use(976) code(926) identifi(902) }
{ drug(1928) target(777) effect(648) }
{ implement(1333) system(1263) develop(1122) }
{ survey(1388) particip(1329) question(1065) }
{ estim(2440) model(1874) function(577) }
{ decis(3086) make(1611) patient(1517) }
{ process(1125) use(805) approach(778) }
{ activ(1452) weight(1219) physic(1104) }
{ method(1969) cluster(1462) data(1082) }
{ detect(2391) sensit(1101) algorithm(908) }

Resumo

An extreme learning machine (ELM) can be regarded as a two-stage feed-forward neural network (FNN) learning system that randomly assigns the connections with and within hidden neurons in the first stage and tunes the connections with output neurons in the second stage. Therefore, ELM training is essentially a linear learning problem, which significantly reduces the computational burden. Numerous applications show that such a computation burden reduction does not degrade the generalization capability. It has, however, been open that whether this is true in theory. The aim of this paper is to study the theoretical feasibility of ELM by analyzing the pros and cons of ELM. In the previous part of this topic, we pointed out that via appropriately selected activation functions, ELM does not degrade the generalization capability in the sense of expectation. In this paper, we launch the study in a different direction and show that the randomness of ELM also leads to certain negative consequences. On one hand, we find that the randomness causes an additional uncertainty problem of ELM, both in approximation and learning. On the other hand, we theoretically justify that there also exist activation functions such that the corresponding ELM degrades the generalization capability. In particular, we prove that the generalization capability of ELM with Gaussian kernel is essentially worse than that of FNN with Gaussian kernel. To facilitate the use of ELM, we also provide a remedy to such a degradation. We find that the well-developed coefficient regularization technique can essentially improve the generalization capability. The obtained results reveal the essential characteristic of ELM in a certain sense and give theoretical guidance concerning how to use ELM.

Resumo Limpo

extrem learn machin elm can regard twostag feedforward neural network fnn learn system random assign connect within hidden neuron first stage tune connect output neuron second stage therefor elm train essenti linear learn problem signific reduc comput burden numer applic show comput burden reduct degrad general capabl howev open whether true theori aim paper studi theoret feasibl elm analyz pros con elm previous part topic point via appropri select activ function elm degrad general capabl sens expect paper launch studi differ direct show random elm also lead certain negat consequ one hand find random caus addit uncertainti problem elm approxim learn hand theoret justifi also exist activ function correspond elm degrad general capabl particular prove general capabl elm gaussian kernel essenti wors fnn gaussian kernel facilit use elm also provid remedi degrad find welldevelop coeffici regular techniqu can essenti improv general capabl obtain result reveal essenti characterist elm certain sens give theoret guidanc concern use elm

Resumos Similares

IEEE Trans Neural Netw Learn Syst - Adaptive neural control of nonlinear MIMO systems with time-varying output constraints. ( 0,659672786449253 )
Neural Comput - A new supervised learning algorithm for spiking neurons. ( 0,657446726568954 )
IEEE Trans Pattern Anal Mach Intell - Iterative Quantization: A Procrustean Approach to Learning Binary Codes for Large-scale Image Retrieval. ( 0,653069691921133 )
IEEE Trans Neural Netw Learn Syst - Passivity and Passification of Memristor-Based Recurrent Neural Networks With Additive Time-Varying Delays. ( 0,652233738047548 )
Comput Methods Programs Biomed - Design of fuzzy classifier for diabetes disease using Modified Artificial Bee Colony algorithm. ( 0,645616997490653 )
IEEE Trans Neural Netw Learn Syst - Scalable Nonparametric Low-Rank Kernel Learning Using Block Coordinate Descent. ( 0,629428942760954 )
IEEE Trans Pattern Anal Mach Intell - Constrained Nonnegative Matrix Factorization for Image Representation. ( 0,628439196209011 )
IEEE Trans Neural Netw Learn Syst - Is extreme learning machine feasible? A theoretical assessment (part I). ( 0,625843714819047 )
Neural Comput - Learning rates of lq coefficient regularization learning with gaussian kernel. ( 0,615570345612883 )
J Biomed Inform - A medical diagnostic tool based on radial basis function classifiers and evolutionary simulated annealing. ( 0,614510367119262 )
Neural Comput - Robust subspace discovery via relaxed rank minimization. ( 0,609885722654056 )
Comput Math Methods Med - Energy preserved sampling for compressed sensing MRI. ( 0,600879149251679 )
Int J Neural Syst - Evolving RBF neural networks for adaptive soft-sensor design. ( 0,599900350551395 )
IEEE Trans Image Process - Deconvolving images with unknown boundaries using the alternating direction method of multipliers. ( 0,599689141505649 )
IEEE Trans Image Process - Compressive sensing via nonlocal low-rank regularization. ( 0,594843295857506 )
Neural Comput - Direct learning of sparse changes in Markov networks by density ratio estimation. ( 0,594822698298728 )
Comput Biol Chem - A hyper-heuristic for the Longest Common Subsequence problem. ( 0,593706488031725 )
IEEE Trans Image Process - Blind deconvolution using generalized cross-validation approach to regularization parameter estimation. ( 0,593445361240264 )
Neural Comput - Encoding binary neural codes in networks of threshold-linear neurons. ( 0,592891122176207 )
IEEE J Biomed Health Inform - Multiple kernel learning in the primal for multimodal Alzheimer's disease classification. ( 0,590518185641496 )
J. Comput. Biol. - The worst case complexity of maximum parsimony. ( 0,588660223964307 )
IEEE Trans Image Process - Efficient algorithms for robust recovery of images from compressed data. ( 0,585704077653923 )
J Biomed Inform - Error-correction learning for artificial neural networks using the Bayesian paradigm. Application to automated medical diagnosis. ( 0,584450350297817 )
IEEE Trans Neural Netw Learn Syst - Further result on guaranteed H8 performance state estimation of delayed static neural networks. ( 0,584144618632635 )
IEEE Trans Image Process - On preserving original variables in Bayesian PCA with application to image analysis. ( 0,584070025665221 )
IEEE Trans Neural Netw Learn Syst - Dynamic Surface Control Using Neural Networks for a Class of Uncertain Nonlinear Systems With Input Saturation. ( 0,584009520228553 )
Neural Comput - Regularized variational Bayesian learning of echo state networks with delay&sum readout. ( 0,583461479740045 )
IEEE Trans Image Process - Coupled dictionary training for image super-resolution. ( 0,583030036176339 )
IEEE Trans Image Process - Gait recognition with shifted energy image and structural feature extraction. ( 0,582992203155583 )
IEEE Trans Image Process - Efficient rate-distortion optimal packetization of embedded bitstreams into independent source packets. ( 0,580562338366193 )
Neural Comput - Inhibition in multiclass classification. ( 0,574788464665603 )
Comput Math Methods Med - Robust synchronization of delayed chaotic FitzHugh-Nagumo neurons under external electrical stimulation. ( 0,573453990989234 )
IEEE J Biomed Health Inform - On the blind recovery of cardiac and respiratory sounds. ( 0,571314281203312 )
Int J Neural Syst - Optimal sparse approximation with integrate and fire neurons. ( 0,569528313331812 )
IEEE Trans Neural Netw Learn Syst - Online Sequential Extreme Learning Machine With Kernels. ( 0,566210143516808 )
IEEE Trans Neural Netw Learn Syst - Actor-critic-based optimal tracking for partially unknown nonlinear discrete-time systems. ( 0,564093878844436 )
Neural Comput - Guaranteed classification via regularized similarity learning. ( 0,56392964065953 )
IEEE Trans Image Process - Optimal PET protection for streaming scalably compressed video streams with limited retransmission based on incomplete feedback. ( 0,56230517103473 )
Neural Comput - Characterizing responses of translation-invariant neurons to natural stimuli: maximally informative invariant dimensions. ( 0,560115027630114 )
IEEE Trans Image Process - On regularized reconstruction of vector fields. ( 0,558617475242988 )
IEEE Trans Pattern Anal Mach Intell - Robust Recovery of Corrupted Low-rank Matrix by Implicit Regularizers. ( 0,557098424915729 )
IEEE Trans Image Process - Image completion by diffusion maps and spectral relaxation. ( 0,556811815869787 )
Neural Comput - Learning with convex loss and indefinite kernels. ( 0,554978234710233 )
IEEE Trans Image Process - Blind spectral unmixing based on sparse nonnegative matrix factorization. ( 0,554863986725085 )
Neural Comput - Recurrent kernel machines: computing with infinite echo state networks. ( 0,553643798906748 )
IEEE Trans Image Process - Coupled variational image decomposition and restoration model for blurred cartoon-plus-texture images with missing pixels. ( 0,552876036872307 )
Int J Neural Syst - Reactive control of epileptiform discharges in realistic computational neuronal models with bistability. ( 0,551291368551351 )
Neural Comput - Solving the distal reward problem with rare correlations. ( 0,551228256714321 )
IEEE Trans Image Process - l2 Restoration of l8-decoded images via soft-decision estimation. ( 0,550463202180199 )
IEEE Trans Neural Netw Learn Syst - Learning With Mixed Hard/Soft Pointwise Constraints. ( 0,549907655064984 )
IEEE Trans Image Process - A comprehensive framework for image inpainting. ( 0,548212868856578 )
Neural Comput - Active subspace: toward scalable low-rank learning. ( 0,54520137084242 )
Neural Comput - Alternating proximal regularized dictionary learning. ( 0,545026154034784 )
Neural Comput - Synchronization and redundancy: implications for robustness of neural learning and decision making. ( 0,544630506183985 )
IEEE Trans Image Process - A majorize-minimize strategy for subspace optimization applied to image restoration. ( 0,544607434296588 )
IEEE Trans Neural Netw Learn Syst - A two-layer recurrent neural network for nonsmooth convex optimization problems. ( 0,544457966483753 )
IEEE Trans Neural Netw Learn Syst - Adaptive NN Control of a Class of Nonlinear Systems With Asymmetric Saturation Actuators. ( 0,544018694710814 )
Neural Comput - Broken symmetries in a location-invariant word recognition network. ( 0,542415115228047 )
IEEE Trans Pattern Anal Mach Intell - Learning with Augmented Features for Supervised and Semi-supervised Heterogeneous Domain Adaptation. ( 0,542288976672287 )
IEEE Trans Neural Netw Learn Syst - Finite-Horizon Approximate Optimal Guaranteed Cost Control of Uncertain Nonlinear Systems With Application to Mars Entry Guidance. ( 0,541661373641171 )
Neural Comput - Short-term memory capacity in networks via the restricted isometry property. ( 0,539966063907084 )
IEEE Trans Neural Netw Learn Syst - A one-class kernel fisher criterion for outlier detection. ( 0,537277311513205 )
Med Biol Eng Comput - Binary optimization for source localization in the inverse problem of ECG. ( 0,537223486081167 )
IEEE Trans Pattern Anal Mach Intell - A Variance Minimization Criterion to Feature Selection Using Laplacian Regularization. ( 0,536139145068192 )
Neural Comput - A spike-timing-based integrated model for pattern recognition. ( 0,535788960481149 )
Comput. Biol. Med. - Hangman BCI: an unsupervised adaptive self-paced Brain-Computer Interface for playing games. ( 0,535632068603241 )
IEEE Trans Image Process - Bits from photons: oversampled image acquisition using binary Poisson statistics. ( 0,533304380983275 )
IEEE Trans Pattern Anal Mach Intell - Consistent Latent Position Estimation and Vertex Classification for Random Dot Product Graphs. ( 0,533297711217675 )
J. Comput. Biol. - Comparing pedigree graphs. ( 0,53310587617262 )
Comput Biol Chem - A computational model for enhancing recombinant Penicillin G Acylase production from Escherichia coli DH5a. ( 0,53291440157519 )
IEEE Trans Image Process - Artistic image analysis using graph-based learning approaches. ( 0,532718141634564 )
Neural Comput - Hidden Markov models for the stimulus-response relationships of multistate neural systems. ( 0,532498025222293 )
IEEE Trans Image Process - Double shrinking sparse dimension reduction. ( 0,532267094387672 )
IEEE Trans Neural Netw Learn Syst - Comparison of l1-Norm SVR and Sparse Coding Algorithms for Linear Regression. ( 0,531539506128512 )
IEEE Trans Neural Netw Learn Syst - Incremental Support Vector Learning for Ordinal Regression. ( 0,530889538838954 )
IEEE Trans Image Process - Wavelet-based compressed sensing using a Gaussian scale mixture model. ( 0,530721123031705 )
IEEE Trans Image Process - Manifold regularized multitask learning for semi-supervised multilabel image classification. ( 0,53069101373589 )
Neural Comput - Attention as reward-driven optimization of sensory processing. ( 0,530440330090236 )
IEEE Trans Image Process - Modified gradient search for level set based image segmentation. ( 0,530382677340583 )
IEEE Trans Image Process - Joint reconstruction of multiview compressed images. ( 0,529053361897368 )
Neural Comput - A low-order model of biological neural networks. ( 0,528934183223517 )
IEEE Trans Image Process - ADART: an adaptive algebraic reconstruction algorithm for discrete tomography. ( 0,528511849197137 )
IEEE Trans Neural Netw Learn Syst - A Neurodynamic Optimization Method for Recovery of Compressive Sensed Signals With Globally Converged Solution Approximating to l0 Minimization. ( 0,527870873519637 )
IEEE Trans Pattern Anal Mach Intell - What Is Optimized in Convex Relaxations for Multi-Label Problems: Connecting Discrete and Continuously-Inspired MAP Inference. ( 0,527806745614177 )
IEEE Trans Image Process - A fast majorize-minimize algorithm for the recovery of sparse and low-rank matrices. ( 0,527554968774084 )
IEEE Trans Neural Netw Learn Syst - Kernel reconstruction ICA for sparse representation. ( 0,526438692238637 )
Neural Comput - A no-go theorem for one-layer feedforward networks. ( 0,526147166759376 )
J Chem Inf Model - Hybrid genetic algorithm with an adaptive penalty function for fitting multimodal experimental data: application to exchange-coupled non-Kramers binuclear iron active sites. ( 0,525842331918103 )
IEEE Trans Image Process - Robust image analysis with sparse representation on quantized visual features. ( 0,525057480653052 )
Neural Comput - Generalization and multirate models of motor adaptation. ( 0,52483619581425 )
IEEE Trans Pattern Anal Mach Intell - Graph Regularized Nonnegative Matrix Factorization for Data Representation. ( 0,523176552594347 )
IEEE Trans Pattern Anal Mach Intell - Automatic Generation of Co-Embeddings from Relational Data with Adaptive Shaping. ( 0,522896011288381 )
Neural Comput - Insights from a simple expression for linear fisher information in a recurrently connected population of spiking neurons. ( 0,522809575051525 )
Neural Comput - Neuronal assembly dynamics in supervised and unsupervised learning scenarios. ( 0,522577837876335 )
IEEE Trans Image Process - A regularized model-based optimization framework for pan-sharpening. ( 0,522118481404205 )
J Biomed Inform - A novel artificial neural network method for biomedical prediction based on matrix pseudo-inversion. ( 0,521303265422024 )
Neural Comput - Decorrelation of spiking variability and improved information transfer through feedforward divisive normalization. ( 0,521100496939486 )
Neural Comput - A proof of convergence of the concave-convex procedure using Zangwill's theory. ( 0,520316047746685 )
IEEE Trans Image Process - Generalized Nash Bargaining Solution to Rate Control Optimization for Spatial Scalable Video Coding. ( 0,520053497819218 )
IEEE Trans Neural Netw Learn Syst - Phase Oscillatory Network and Visual Pattern Recognition. ( 0,519615309215875 )