-
Wolfekuhler, M.R.; Punch, W.F.: Finding salient features for personal Web pages categories (1997)
0.04
0.041996814 = product of:
0.12599044 = sum of:
0.12599044 = sum of:
0.08138241 = weight(_text_:networks in 2673) [ClassicSimilarity], result of:
0.08138241 = score(doc=2673,freq=2.0), product of:
0.22247115 = queryWeight, product of:
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.047034867 = queryNorm
0.36581108 = fieldWeight in 2673, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.0546875 = fieldNorm(doc=2673)
0.044608027 = weight(_text_:22 in 2673) [ClassicSimilarity], result of:
0.044608027 = score(doc=2673,freq=2.0), product of:
0.1647081 = queryWeight, product of:
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.047034867 = queryNorm
0.2708308 = fieldWeight in 2673, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.0546875 = fieldNorm(doc=2673)
0.33333334 = coord(1/3)
- Date
- 1. 8.1996 22:08:06
- Source
- Computer networks and ISDN systems. 29(1997) no.8, S.1147-1156
-
Mao, J.; Xu, W.; Yang, Y.; Wang, J.; Yuille, A.L.: Explain images with multimodal recurrent neural networks (2014)
0.02
0.020136926 = product of:
0.060410775 = sum of:
0.060410775 = product of:
0.12082155 = sum of:
0.12082155 = weight(_text_:networks in 1557) [ClassicSimilarity], result of:
0.12082155 = score(doc=1557,freq=6.0), product of:
0.22247115 = queryWeight, product of:
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.047034867 = queryNorm
0.5430886 = fieldWeight in 1557, product of:
2.4494898 = tf(freq=6.0), with freq of:
6.0 = termFreq=6.0
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.046875 = fieldNorm(doc=1557)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Abstract
- In this paper, we present a multimodal Recurrent Neural Network (m-RNN) model for generating novel sentence descriptions to explain the content of images. It directly models the probability distribution of generating a word given previous words and the image. Image descriptions are generated by sampling from this distribution. The model consists of two sub-networks: a deep recurrent neural network for sentences and a deep convolutional network for images. These two sub-networks interact with each other in a multimodal layer to form the whole m-RNN model. The effectiveness of our model is validated on three benchmark datasets (IAPR TC-12 [8], Flickr 8K [28], and Flickr 30K [13]). Our model outperforms the state-of-the-art generative method. In addition, the m-RNN model can be applied to retrieval tasks for retrieving images or sentences, and achieves significant performance improvement over the state-of-the-art methods which directly optimize the ranking objective function for retrieval.
-
Krutulis, J.D.; Jacob, E.K.: ¬A theoretical model for the study of emergent structure in adaptive information networks (1995)
0.02
0.019182019 = product of:
0.057546057 = sum of:
0.057546057 = product of:
0.11509211 = sum of:
0.11509211 = weight(_text_:networks in 3353) [ClassicSimilarity], result of:
0.11509211 = score(doc=3353,freq=4.0), product of:
0.22247115 = queryWeight, product of:
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.047034867 = queryNorm
0.517335 = fieldWeight in 3353, product of:
2.0 = tf(freq=4.0), with freq of:
4.0 = termFreq=4.0
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.0546875 = fieldNorm(doc=3353)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Abstract
- Attempts to automate classification have focused on mimicking the intellectual processes whereby human classifiers assign entities to mutually exclusive groups that exhibit or more shared characteristics. A more viable approach might be to construct an adaptive retrieval system that produces groupings of related entities by generating dynamic categories based on document content and on the system's emergent structure as it adapts to modifications in the database and to observed patterns of access. Presents a theoretical model for adaptive information networks using relevance feedback and genetic algorithms to generate emergent structure
-
Voorhees, E.M.: Implementing agglomerative hierarchic clustering algorithms for use in document retrieval (1986)
0.02
0.016993534 = product of:
0.0509806 = sum of:
0.0509806 = product of:
0.1019612 = sum of:
0.1019612 = weight(_text_:22 in 402) [ClassicSimilarity], result of:
0.1019612 = score(doc=402,freq=2.0), product of:
0.1647081 = queryWeight, product of:
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.047034867 = queryNorm
0.61904186 = fieldWeight in 402, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.125 = fieldNorm(doc=402)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Source
- Information processing and management. 22(1986) no.6, S.465-476
-
Ma, N.; Zheng, H.T.; Xiao, X.: ¬An ontology-based latent semantic indexing approach using long short-term memory networks (2017)
0.02
0.016780771 = product of:
0.050342314 = sum of:
0.050342314 = product of:
0.10068463 = sum of:
0.10068463 = weight(_text_:networks in 3810) [ClassicSimilarity], result of:
0.10068463 = score(doc=3810,freq=6.0), product of:
0.22247115 = queryWeight, product of:
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.047034867 = queryNorm
0.45257387 = fieldWeight in 3810, product of:
2.4494898 = tf(freq=6.0), with freq of:
6.0 = termFreq=6.0
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.0390625 = fieldNorm(doc=3810)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Abstract
- Nowadays, online data shows an astonishing increase and the issue of semantic indexing remains an open question. Ontologies and knowledge bases have been widely used to optimize performance. However, researchers are placing increased emphasis on internal relations of ontologies but neglect latent semantic relations between ontologies and documents. They generally annotate instances mentioned in documents, which are related to concepts in ontologies. In this paper, we propose an Ontology-based Latent Semantic Indexing approach utilizing Long Short-Term Memory networks (LSTM-OLSI). We utilize an importance-aware topic model to extract document-level semantic features and leverage ontologies to extract word-level contextual features. Then we encode the above two levels of features and match their embedding vectors utilizing LSTM networks. Finally, the experimental results reveal that LSTM-OLSI outperforms existing techniques and demonstrates deep comprehension of instances and articles.
-
Karpathy, A.; Fei-Fei, L.: Deep visual-semantic alignments for generating image descriptions (2015)
0.02
0.01644173 = product of:
0.04932519 = sum of:
0.04932519 = product of:
0.09865038 = sum of:
0.09865038 = weight(_text_:networks in 1868) [ClassicSimilarity], result of:
0.09865038 = score(doc=1868,freq=4.0), product of:
0.22247115 = queryWeight, product of:
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.047034867 = queryNorm
0.44343 = fieldWeight in 1868, product of:
2.0 = tf(freq=4.0), with freq of:
4.0 = termFreq=4.0
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.046875 = fieldNorm(doc=1868)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Abstract
- We present a model that generates free-form natural language descriptions of image regions. Our model leverages datasets of images and their sentence descriptions to learn about the inter-modal correspondences between text and visual data. Our approach is based on a novel combination of Convolutional Neural Networks over image regions, bidirectional Recurrent Neural Networks over sentences, and a structured objective that aligns the two modalities through a multimodal embedding. We then describe a Recurrent Neural Network architecture that uses the inferred alignments to learn to generate novel descriptions of image regions. We demonstrate the effectiveness of our alignment model with ranking experiments on Flickr8K, Flickr30K and COCO datasets, where we substantially improve on the state of the art. We then show that the sentences created by our generative model outperform retrieval baselines on the three aforementioned datasets and a new dataset of region-level annotations.
-
Fuhr, N.; Niewelt, B.: ¬Ein Retrievaltest mit automatisch indexierten Dokumenten (1984)
0.01
0.014869343 = product of:
0.044608027 = sum of:
0.044608027 = product of:
0.08921605 = sum of:
0.08921605 = weight(_text_:22 in 262) [ClassicSimilarity], result of:
0.08921605 = score(doc=262,freq=2.0), product of:
0.1647081 = queryWeight, product of:
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.047034867 = queryNorm
0.5416616 = fieldWeight in 262, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.109375 = fieldNorm(doc=262)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Date
- 20.10.2000 12:22:23
-
Hlava, M.M.K.: Automatic indexing : comparing rule-based and statistics-based indexing systems (2005)
0.01
0.014869343 = product of:
0.044608027 = sum of:
0.044608027 = product of:
0.08921605 = sum of:
0.08921605 = weight(_text_:22 in 6265) [ClassicSimilarity], result of:
0.08921605 = score(doc=6265,freq=2.0), product of:
0.1647081 = queryWeight, product of:
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.047034867 = queryNorm
0.5416616 = fieldWeight in 6265, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.109375 = fieldNorm(doc=6265)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Source
- Information outlook. 9(2005) no.8, S.22-23
-
Donahue, J.; Hendricks, L.A.; Guadarrama, S.; Rohrbach, M.; Venugopalan, S.; Saenko, K.; Darrell, T.: Long-term recurrent convolutional networks for visual recognition and description (2014)
0.01
0.013701442 = product of:
0.041104324 = sum of:
0.041104324 = product of:
0.08220865 = sum of:
0.08220865 = weight(_text_:networks in 1873) [ClassicSimilarity], result of:
0.08220865 = score(doc=1873,freq=4.0), product of:
0.22247115 = queryWeight, product of:
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.047034867 = queryNorm
0.369525 = fieldWeight in 1873, product of:
2.0 = tf(freq=4.0), with freq of:
4.0 = termFreq=4.0
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.0390625 = fieldNorm(doc=1873)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Abstract
- Models based on deep convolutional networks have dominated recent image interpretation tasks; we investigate whether models which are also recurrent, or "temporally deep", are effective for tasks involving sequences, visual and otherwise. We develop a novel recurrent convolutional architecture suitable for large-scale visual learning which is end-to-end trainable, and demonstrate the value of these models on benchmark video recognition tasks, image description and retrieval problems, and video narration challenges. In contrast to current models which assume a fixed spatio-temporal receptive field or simple temporal averaging for sequential processing, recurrent convolutional models are "doubly deep" in that they can be compositional in spatial and temporal "layers". Such models may have advantages when target concepts are complex and/or training data are limited. Learning long-term dependencies is possible when nonlinearities are incorporated into the network state updates. Long-term RNN models are appealing in that they directly can map variable-length inputs (e.g., video frames) to variable length outputs (e.g., natural language text) and can model complex temporal dynamics; yet they can be optimized with backpropagation. Our recurrent long-term models are directly connected to modern visual convnet models and can be jointly trained to simultaneously learn temporal dynamics and convolutional perceptual representations. Our results show such models have distinct advantages over state-of-the-art models for recognition or generation which are separately defined and/or optimized.
-
Pfeifer, U.; Fuhr, N.; Huynh, T.: Searching structured documents with the enhanced retrieval functionality of freeWAIS-sf and SFgate (1995)
0.01
0.013563735 = product of:
0.040691204 = sum of:
0.040691204 = product of:
0.08138241 = sum of:
0.08138241 = weight(_text_:networks in 2214) [ClassicSimilarity], result of:
0.08138241 = score(doc=2214,freq=2.0), product of:
0.22247115 = queryWeight, product of:
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.047034867 = queryNorm
0.36581108 = fieldWeight in 2214, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.0546875 = fieldNorm(doc=2214)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Source
- Computer networks and ISDN systems. 27(1995) no.6, S.1027-36
-
Fuhr, N.: Ranking-Experimente mit gewichteter Indexierung (1986)
0.01
0.012745151 = product of:
0.038235452 = sum of:
0.038235452 = product of:
0.076470904 = sum of:
0.076470904 = weight(_text_:22 in 58) [ClassicSimilarity], result of:
0.076470904 = score(doc=58,freq=2.0), product of:
0.1647081 = queryWeight, product of:
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.047034867 = queryNorm
0.46428138 = fieldWeight in 58, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.09375 = fieldNorm(doc=58)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Date
- 14. 6.2015 22:12:44
-
Hauer, M.: Automatische Indexierung (2000)
0.01
0.012745151 = product of:
0.038235452 = sum of:
0.038235452 = product of:
0.076470904 = sum of:
0.076470904 = weight(_text_:22 in 5887) [ClassicSimilarity], result of:
0.076470904 = score(doc=5887,freq=2.0), product of:
0.1647081 = queryWeight, product of:
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.047034867 = queryNorm
0.46428138 = fieldWeight in 5887, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.09375 = fieldNorm(doc=5887)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Source
- Wissen in Aktion: Wege des Knowledge Managements. 22. Online-Tagung der DGI, Frankfurt am Main, 2.-4.5.2000. Proceedings. Hrsg.: R. Schmidt
-
Fuhr, N.: Rankingexperimente mit gewichteter Indexierung (1986)
0.01
0.012745151 = product of:
0.038235452 = sum of:
0.038235452 = product of:
0.076470904 = sum of:
0.076470904 = weight(_text_:22 in 2051) [ClassicSimilarity], result of:
0.076470904 = score(doc=2051,freq=2.0), product of:
0.1647081 = queryWeight, product of:
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.047034867 = queryNorm
0.46428138 = fieldWeight in 2051, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.09375 = fieldNorm(doc=2051)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Date
- 14. 6.2015 22:12:56
-
Hauer, M.: Tiefenindexierung im Bibliothekskatalog : 17 Jahre intelligentCAPTURE (2019)
0.01
0.012745151 = product of:
0.038235452 = sum of:
0.038235452 = product of:
0.076470904 = sum of:
0.076470904 = weight(_text_:22 in 5629) [ClassicSimilarity], result of:
0.076470904 = score(doc=5629,freq=2.0), product of:
0.1647081 = queryWeight, product of:
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.047034867 = queryNorm
0.46428138 = fieldWeight in 5629, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.09375 = fieldNorm(doc=5629)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Source
- B.I.T.online. 22(2019) H.2, S.163-166
-
Biebricher, N.; Fuhr, N.; Lustig, G.; Schwantner, M.; Knorz, G.: ¬The automatic indexing system AIR/PHYS : from research to application (1988)
0.01
0.010620959 = product of:
0.031862877 = sum of:
0.031862877 = product of:
0.063725755 = sum of:
0.063725755 = weight(_text_:22 in 1952) [ClassicSimilarity], result of:
0.063725755 = score(doc=1952,freq=2.0), product of:
0.1647081 = queryWeight, product of:
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.047034867 = queryNorm
0.38690117 = fieldWeight in 1952, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.078125 = fieldNorm(doc=1952)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Date
- 16. 8.1998 12:51:22
-
Kutschekmanesch, S.; Lutes, B.; Moelle, K.; Thiel, U.; Tzeras, K.: Automated multilingual indexing : a synthesis of rule-based and thesaurus-based methods (1998)
0.01
0.010620959 = product of:
0.031862877 = sum of:
0.031862877 = product of:
0.063725755 = sum of:
0.063725755 = weight(_text_:22 in 4157) [ClassicSimilarity], result of:
0.063725755 = score(doc=4157,freq=2.0), product of:
0.1647081 = queryWeight, product of:
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.047034867 = queryNorm
0.38690117 = fieldWeight in 4157, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.078125 = fieldNorm(doc=4157)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Source
- Information und Märkte: 50. Deutscher Dokumentartag 1998, Kongreß der Deutschen Gesellschaft für Dokumentation e.V. (DGD), Rheinische Friedrich-Wilhelms-Universität Bonn, 22.-24. September 1998. Hrsg. von Marlies Ockenfeld u. Gerhard J. Mantwill
-
Tsareva, P.V.: Algoritmy dlya raspoznavaniya pozitivnykh i negativnykh vkhozdenii deskriptorov v tekst i protsedura avtomaticheskoi klassifikatsii tekstov (1999)
0.01
0.010620959 = product of:
0.031862877 = sum of:
0.031862877 = product of:
0.063725755 = sum of:
0.063725755 = weight(_text_:22 in 374) [ClassicSimilarity], result of:
0.063725755 = score(doc=374,freq=2.0), product of:
0.1647081 = queryWeight, product of:
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.047034867 = queryNorm
0.38690117 = fieldWeight in 374, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.078125 = fieldNorm(doc=374)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Date
- 1. 4.2002 10:22:41
-
Stankovic, R. et al.: Indexing of textual databases based on lexical resources : a case study for Serbian (2016)
0.01
0.010620959 = product of:
0.031862877 = sum of:
0.031862877 = product of:
0.063725755 = sum of:
0.063725755 = weight(_text_:22 in 2759) [ClassicSimilarity], result of:
0.063725755 = score(doc=2759,freq=2.0), product of:
0.1647081 = queryWeight, product of:
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.047034867 = queryNorm
0.38690117 = fieldWeight in 2759, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
3.5018296 = idf(docFreq=3622, maxDocs=44218)
0.078125 = fieldNorm(doc=2759)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Date
- 1. 2.2016 18:25:22
-
Moreno, J.M.T.: Automatic text summarization (2014)
0.01
0.009688382 = product of:
0.029065145 = sum of:
0.029065145 = product of:
0.05813029 = sum of:
0.05813029 = weight(_text_:networks in 1518) [ClassicSimilarity], result of:
0.05813029 = score(doc=1518,freq=2.0), product of:
0.22247115 = queryWeight, product of:
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.047034867 = queryNorm
0.26129362 = fieldWeight in 1518, product of:
1.4142135 = tf(freq=2.0), with freq of:
2.0 = termFreq=2.0
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.0390625 = fieldNorm(doc=1518)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Abstract
- This new textbook examines the motivations and the different algorithms for automatic document summarization (ADS). We performed a recent state of the art. The book shows the main problems of ADS, difficulties and the solutions provided by the community. It presents recent advances in ADS, as well as current applications and trends. The approaches are statistical, linguistic and symbolic. Several exemples are included in order to clarify the theoretical concepts. The books currently available in the area of Automatic Document Summarization are not recent. Powerful algorithms have been developed in recent years that include several applications of ADS. The development of recent technology has impacted on the development of algorithms and their applications. The massive use of social networks and the new forms of the technology requires the adaptation of the classical methods of text summarizers. This is a new textbook on Automatic Text Summarization, based on teaching materials used in two or one-semester courses. It presents a extensive state-of-art and describes the new systems on the subject. Previous automatic summarization books have been either collections of specialized papers, or else authored books with only a chapter or two devoted to the field as a whole. In other hand, the classic books on the subject are not recent.
-
Markoff, J.: Researchers announce advance in image-recognition software (2014)
0.01
0.009688382 = product of:
0.029065145 = sum of:
0.029065145 = product of:
0.05813029 = sum of:
0.05813029 = weight(_text_:networks in 1875) [ClassicSimilarity], result of:
0.05813029 = score(doc=1875,freq=8.0), product of:
0.22247115 = queryWeight, product of:
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.047034867 = queryNorm
0.26129362 = fieldWeight in 1875, product of:
2.828427 = tf(freq=8.0), with freq of:
8.0 = termFreq=8.0
4.72992 = idf(docFreq=1060, maxDocs=44218)
0.01953125 = fieldNorm(doc=1875)
0.5 = coord(1/2)
0.33333334 = coord(1/3)
- Content
- In the longer term, the new research may lead to technology that helps the blind and robots navigate natural environments. But it also raises chilling possibilities for surveillance. During the past 15 years, video cameras have been placed in a vast number of public and private spaces. In the future, the software operating the cameras will not only be able to identify particular humans via facial recognition, experts say, but also identify certain types of behavior, perhaps even automatically alerting authorities. Two years ago Google researchers created image-recognition software and presented it with 10 million images taken from YouTube videos. Without human guidance, the program trained itself to recognize cats - a testament to the number of cat videos on YouTube. Current artificial intelligence programs in new cars already can identify pedestrians and bicyclists from cameras positioned atop the windshield and can stop the car automatically if the driver does not take action to avoid a collision. But "just single object recognition is not very beneficial," said Ali Farhadi, a computer scientist at the University of Washington who has published research on software that generates sentences from digital pictures. "We've focused on objects, and we've ignored verbs," he said, adding that these programs do not grasp what is going on in an image. Both the Google and Stanford groups tackled the problem by refining software programs known as neural networks, inspired by our understanding of how the brain works. Neural networks can "train" themselves to discover similarities and patterns in data, even when their human creators do not know the patterns exist.
In living organisms, webs of neurons in the brain vastly outperform even the best computer-based networks in perception and pattern recognition. But by adopting some of the same architecture, computers are catching up, learning to identify patterns in speech and imagery with increasing accuracy. The advances are apparent to consumers who use Apple's Siri personal assistant, for example, or Google's image search. Both groups of researchers employed similar approaches, weaving together two types of neural networks, one focused on recognizing images and the other on human language. In both cases the researchers trained the software with relatively small sets of digital images that had been annotated with descriptive sentences by humans. After the software programs "learned" to see patterns in the pictures and description, the researchers turned them on previously unseen images. The programs were able to identify objects and actions with roughly double the accuracy of earlier efforts, although still nowhere near human perception capabilities. "I was amazed that even with the small amount of training data that we were able to do so well," said Oriol Vinyals, a Google computer scientist who wrote the paper with Alexander Toshev, Samy Bengio and Dumitru Erhan, members of the Google Brain project. "The field is just starting, and we will see a lot of increases."
