IEEE Trans Pattern Anal Mach Intell - Good Practice in Large-Scale Learning for Image Classification.

Tópicos

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

Resumo

We benchmark several SVM objective functions for large-scale image classification. We consider one-vs-rest, multi-class, ranking, and weighted approximate ranking SVMs. A comparison of online and batch methods for optimizing the objectives shows that online methods perform as well as batch methods in terms of classification accuracy, but with a significant gain in training speed. Using stochastic gradient descent, we can scale the training to millions of images and thousands of classes. Our experimental evaluation shows that ranking-based algorithms do not outperform the one-vs-rest strategy when a large number of training examples are used. Furthermore, the gap in accuracy between the different algorithms shrinks as the dimension of the features increases. We also show that learning through cross-validation the optimal rebalancing of positive and negative examples can result in a significant improvement for the one-vs-rest strategy. Finally, early stopping can be used as an effective regularization strategy when training with online algorithms. Following these "good practices", we were able to improve the state-of-the-art on a large subset of 10K classes and 9M images of ImageNet from 16.7% Top-1 accuracy to 19.1%.

Resumo Limpo

benchmark sever svm object function largescal imag classif consid onevsrest multiclass rank weight approxim rank svms comparison onlin batch method optim object show onlin method perform well batch method term classif accuraci signific gain train speed use stochast gradient descent can scale train million imag thousand class experiment evalu show rankingbas algorithm outperform onevsrest strategi larg number train exampl use furthermor gap accuraci differ algorithm shrink dimens featur increas also show learn crossvalid optim rebalanc posit negat exampl can result signific improv onevsrest strategi final earli stop can use effect regular strategi train onlin algorithm follow good practic abl improv stateoftheart larg subset k class m imag imagenet top accuraci

Resumos Similares

IEEE Trans Pattern Anal Mach Intell - Distance-Based Image Classification: Generalizing to New Classes at Near Zero Cost. ( 0,780451105902677 )
Artif Intell Med - A fuzzy-based data transformation for feature extraction to increase classification performance with small medical data sets. ( 0,759896674427639 )
IEEE Trans Pattern Anal Mach Intell - Latent Dirichlet Allocation Models for Image Classification. ( 0,757068126087082 )
Int J Neural Syst - Aggregation of sparse linear discriminant analyses for event-related potential classification in brain-computer interface. ( 0,732266511776561 )
IEEE Trans Image Process - Task-specific image partitioning. ( 0,722476115222869 )
IEEE Trans Image Process - Manifold regularized multitask learning for semi-supervised multilabel image classification. ( 0,718242493529291 )
Comput. Biol. Med. - Robust prediction of protein subcellular localization combining PCA and WSVMs. ( 0,716469668722583 )
Neural Comput - Large margin low rank tensor analysis. ( 0,713997057492015 )
Int J Neural Syst - Structurally enhanced incremental neural learning for image classification with subgraph extraction. ( 0,713393709404728 )
Comput Math Methods Med - On multilabel classification methods of incompletely labeled biomedical text data. ( 0,707436166615441 )
IEEE Trans Image Process - Geodesic propagation for semantic labeling. ( 0,701897019933189 )
Neural Comput - Reduction from cost-sensitive ordinal ranking to weighted binary classification. ( 0,701287030886737 )
IEEE Trans Pattern Anal Mach Intell - Learning Hierarchical Features for Scene Labeling. ( 0,700742666229385 )
IEEE Trans Pattern Anal Mach Intell - Label Consistent K-SVD: Learning A Discriminative Dictionary for Recognition. ( 0,700605790205871 )
Comput Methods Programs Biomed - Modified CC-LR algorithm with three diverse feature sets for motor imagery tasks classification in EEG based brain-computer interface. ( 0,694627989638323 )
IEEE Trans Pattern Anal Mach Intell - Feature Selection and Kernel Learning for Local Learning-Based Clustering. ( 0,675636294453944 )
IEEE Trans Image Process - Grassmannian regularized structured multi-view embedding for image classification. ( 0,671369408996801 )
IEEE Trans Image Process - Active learning for solving the incomplete data problem in facial age classification by the furthest nearest-neighbor criterion. ( 0,669605840385526 )
Comput. Biol. Med. - Sparse Manifold Clustering and Embedding to discriminate gene expression profiles of glioblastoma and meningioma tumors. ( 0,66913825619382 )
IEEE Trans Neural Netw Learn Syst - A Kernel Classification Framework for Metric Learning. ( 0,666430238676137 )
IEEE Trans Image Process - A linear support higher-order tensor machine for classification. ( 0,66567293187919 )
IEEE Trans Image Process - Structured max-margin learning for inter-related classifier training and multilabel image annotation. ( 0,662179570274863 )
Neural Comput - Online learning with (multiple) kernels: a review. ( 0,660265644919878 )
J Chem Inf Model - Note on naive Bayes based on binary descriptors in cheminformatics. ( 0,659224777224229 )
Comput Methods Programs Biomed - Auto-adaptive robot-aided therapy using machine learning techniques. ( 0,658657006434747 )
IEEE Trans Neural Netw Learn Syst - Adaptive Batch Mode Active Learning. ( 0,657055543655053 )
IEEE Trans Pattern Anal Mach Intell - A Bag-of-Features Framework to Classify Time Series. ( 0,655667401021829 )
Int J Neural Syst - Online semi-supervised growing neural gas. ( 0,654607471568275 )
IEEE Trans Image Process - Joint segmentation of images and scanned point cloud in large-scale street scenes with low-annotation cost. ( 0,654358732245159 )
IEEE Trans Image Process - Improving Web image search by bag-based reranking. ( 0,654033831209006 )
IEEE Trans Image Process - Design of non-linear kernel dictionaries for object recognition. ( 0,652617238205439 )
J Chem Inf Model - Classifying large chemical data sets: using a regularized potential function method. ( 0,64530017152097 )
Neural Comput - Adaptive metric learning vector quantization for ordinal classification. ( 0,644206525952141 )
Comput Math Methods Med - Correlation kernels for support vector machines classification with applications in cancer data. ( 0,641988145154995 )
Neural Comput - Divergence-based vector quantization. ( 0,640914515693129 )
IEEE Trans Image Process - Multiple-kernel, multiple-instance similarity features for efficient visual object detection. ( 0,64028498376898 )
IEEE Trans Image Process - An online learning approach to occlusion boundary detection. ( 0,639375048750333 )
J Biomed Inform - Semi-supervised clinical text classification with Laplacian SVMs: an application to cancer case management. ( 0,639213986492253 )
Neural Comput - Computing sparse representations of multidimensional signals using Kronecker bases. ( 0,639171307032334 )
IEEE Trans Image Process - Hyperspectral image classification through bilayer graph-based learning. ( 0,639067739445518 )
IEEE Trans Image Process - Local learned dictionaries optimized to edge orientation for inverse halftoning. ( 0,63179666624949 )
J Med Syst - 3D similarity-dissimilarity plot for high dimensional data visualization in the context of biomedical pattern classification. ( 0,629543165598974 )
IEEE Trans Pattern Anal Mach Intell - Facial Age Estimation by Learning from Label Distributions. ( 0,627357763344313 )
J Am Med Inform Assoc - Learning classification models with soft-label information. ( 0,626819846469877 )
Artif Intell Med - A classifier ensemble approach for the missing feature problem. ( 0,626713740511713 )
IEEE Trans Pattern Anal Mach Intell - Trainable Convolution Filters and Their Application to Face Recognition. ( 0,625684735320128 )
J Biomed Inform - Portable automatic text classification for adverse drug reaction detection via multi-corpus training. ( 0,623596348097424 )
IEEE Trans Pattern Anal Mach Intell - Feature Selection with Conjunctions of Decision Stumps and Learning from Microarray Data. ( 0,620019262671983 )
J Biomed Inform - Multi-label classification of chronically ill patients with bag of words and supervised dimensionality reduction algorithms. ( 0,61916558032414 )
J Biomed Inform - Preparing an annotated gold standard corpus to share with extramural investigators for de-identification research. ( 0,618605558891282 )
J Biomed Inform - Class proximity measures--dissimilarity-based classification and display of high-dimensional data. ( 0,617821551931903 )
IEEE Trans Image Process - Incremental training of a detector using online sparse eigendecomposition. ( 0,61755968694993 )
Neural Comput - Adaptive multiclass classification for brain computer interfaces. ( 0,617144017615103 )
J Biomed Inform - Incremental Gaussian Discriminant Analysis based on Graybill and Deal weighted combination of estimators for brain tumour diagnosis. ( 0,615939607831693 )
BMC Med Inform Decis Mak - Decision tree-based learning to predict patient controlled analgesia consumption and readjustment. ( 0,614567962866943 )
J. Comput. Biol. - Imbalanced class learning in epigenetics. ( 0,613489658503134 )
Comput Math Methods Med - A novel multiinstance learning approach for liver cancer recognition on abdominal CT images based on CPSO-SVM and IO. ( 0,611798832984637 )
Comput Methods Programs Biomed - Computer-aided diagnosis system: a Bayesian hybrid classification method. ( 0,610336697720447 )
IEEE Trans Image Process - Unsupervised amplitude and texture classification of SAR images with multinomial latent model. ( 0,61033280584788 )
IEEE Trans Image Process - Cooperative sparse representation in two opposite directions for semi-supervised image annotation. ( 0,610128893213369 )
Neural Comput - Metacognitive learning in a fully complex-valued radial basis function neural network. ( 0,606833671974335 )
Int J Neural Syst - Span: spike pattern association neuron for learning spatio-temporal spike patterns. ( 0,606823811790289 )
Neural Comput - On the relation of slow feature analysis and Laplacian eigenmaps. ( 0,606098379598148 )
IEEE Trans Image Process - Enhancing training collections for image annotation: an instance-weighted mixture modeling approach. ( 0,605507221231713 )
AMIA Annu Symp Proc - Learning medical diagnosis models from multiple experts. ( 0,605308780709872 )
IEEE Trans Image Process - Multiview Hessian regularization for image annotation. ( 0,60432461183866 )
Comput. Biol. Med. - Automated Marsh-like classification of celiac disease in children using local texture operators. ( 0,603843739854702 )
IEEE Trans Image Process - Artistic image analysis using graph-based learning approaches. ( 0,602676737837592 )
IEEE Trans Neural Netw Learn Syst - ML-Tree: a tree-structure-based approach to multilabel learning. ( 0,601870791704946 )
Comput. Biol. Med. - Relabeling algorithm for retrieval of noisy instances and improving prediction quality. ( 0,60143017884292 )
IEEE J Biomed Health Inform - Systematic Poisoning Attacks on and Defenses for Machine Learning in Healthcare. ( 0,599865757251148 )
IEEE Trans Neural Netw Learn Syst - An efficient topological distance-based tree kernel. ( 0,59907746379184 )
IEEE Trans Image Process - Self-supervised online metric learning with low rank constraint for scene categorization. ( 0,595888009502122 )
IEEE Trans Image Process - Contextual kernel and spectral methods for learning the semantics of images. ( 0,591547110390898 )
IEEE Trans Image Process - Object detection with DoG scale-space: a multiple kernel learning approach. ( 0,590960929538845 )
J Biomed Inform - Learning classification models from multiple experts. ( 0,59069015717033 )
IEEE Trans Pattern Anal Mach Intell - The Effect of Model Misspecification on Semi-Supervised Classification. ( 0,588874063589058 )
IEEE Trans Image Process - Discriminative shared Gaussian processes for multiview and view-invariant facial expression recognition. ( 0,588764093722511 )
J Chem Inf Model - Training based on ligand efficiency improves prediction of bioactivities of ligands and drug target proteins in a machine learning approach. ( 0,586959775051857 )
J. Comput. Biol. - Locally learning biomedical data using diffusion frames. ( 0,584833361938126 )
IEEE Trans Image Process - Learning discriminative dictionary for group sparse representation. ( 0,583870015880785 )
IEEE Trans Pattern Anal Mach Intell - Learning Categories from Few Examples with Multi Model Knowledge Transfer. ( 0,583761703462498 )
Comput. Biol. Med. - A statistical based feature extraction method for breast cancer diagnosis in digital mammogram using multiresolution representation. ( 0,582406953496385 )
BMC Med Inform Decis Mak - Learning to improve medical decision making from imbalanced data without a priori cost. ( 0,582314929465923 )
J Am Med Inform Assoc - Missing values in deduplication of electronic patient data. ( 0,582215612064887 )
IEEE Trans Image Process - Stable orthogonal local discriminant embedding for linear dimensionality reduction. ( 0,581962897832572 )
J Biomed Inform - Applying active learning to assertion classification of concepts in clinical text. ( 0,581754723332932 )
Comput. Biol. Med. - EEG-based emotion estimation using Bayesian weighted-log-posterior function and perceptron convergence algorithm. ( 0,581007294050447 )
IEEE Trans Image Process - Data-dependent hashing based on p-stable distribution. ( 0,58023878584462 )
J Am Med Inform Assoc - Active learning for clinical text classification: is it better than random sampling? ( 0,578729329208724 )
IEEE Trans Image Process - Reducing the complexity of the N-FINDR algorithm for hyperspectral image analysis. ( 0,578034761888922 )
Int J Neural Syst - Linear time relational prototype based learning. ( 0,577277860739321 )
Comput. Biol. Med. - A learning method for the class imbalance problem with medical data sets. ( 0,577125809466581 )
Neural Comput - Unsupervised learning of generative and discriminative weights encoding elementary image components in a predictive coding model of cortical function. ( 0,576085406136819 )
IEEE Trans Neural Netw Learn Syst - Ordinal Distance Metric Learning for Image Ranking. ( 0,572194844491884 )
J Am Med Inform Assoc - Supervised machine learning and active learning in classification of radiology reports. ( 0,571641813361163 )
IEEE Trans Pattern Anal Mach Intell - Weakly Supervised Recognition of Daily Life Activities with Wearable Sensors. ( 0,5712592312278 )
Neural Comput - Incremental learning by message passing in hierarchical temporal memory. ( 0,570121972609056 )
IEEE Trans Pattern Anal Mach Intell - Unsupervised Adaptation Across Domain Shifts By Generating Intermediate Data Representations. ( 0,569611003649018 )
IEEE Trans Image Process - Task-dependent visual-codebook compression. ( 0,569249387399169 )