Search (31 results, page 1 of 2)

0.017291464 = product of:
  0.034582928 = sum of:
    0.034582928 = sum of:
      0.0033826875 = weight(_text_:a in 3780) [ClassicSimilarity], result of:
        0.0033826875 = score(doc=3780,freq=2.0), product of:
          0.053105544 = queryWeight, product of:
            1.153047 = idf(docFreq=37942, maxDocs=44218)
            0.046056706 = queryNorm
          0.06369744 = fieldWeight in 3780, product of:
            1.4142135 = tf(freq=2.0), with freq of:
              2.0 = termFreq=2.0
            1.153047 = idf(docFreq=37942, maxDocs=44218)
            0.0390625 = fieldNorm(doc=3780)
      0.03120024 = weight(_text_:22 in 3780) [ClassicSimilarity], result of:
        0.03120024 = score(doc=3780,freq=2.0), product of:
          0.16128273 = queryWeight, product of:
            3.5018296 = idf(docFreq=3622, maxDocs=44218)
            0.046056706 = queryNorm
          0.19345059 = fieldWeight in 3780, product of:
            1.4142135 = tf(freq=2.0), with freq of:
              2.0 = termFreq=2.0
            3.5018296 = idf(docFreq=3622, maxDocs=44218)
            0.0390625 = fieldNorm(doc=3780)
  0.5 = coord(1/2)

Date

19. 8.2017 9:24:22

Type

a

0.0030184684 = product of:
  0.006036937 = sum of:
    0.006036937 = product of:
      0.012073874 = sum of:
        0.012073874 = weight(_text_:a in 1874) [ClassicSimilarity], result of:
          0.012073874 = score(doc=1874,freq=52.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.22735618 = fieldWeight in 1874, product of:
              7.2111025 = tf(freq=52.0), with freq of:
                52.0 = termFreq=52.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.02734375 = fieldNorm(doc=1874)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Content

"People can summarize a complex scene in a few words without thinking twice. It's much more difficult for computers. But we've just gotten a bit closer -- we've developed a machine-learning system that can automatically produce captions (like the three above) to accurately describe images the first time it sees them. This kind of system could eventually help visually impaired people understand pictures, provide alternate text for images in parts of the world where mobile connections are slow, and make it easier for everyone to search on Google for images. Recent research has greatly improved object detection, classification, and labeling. But accurately describing a complex scene requires a deeper representation of what's going on in the scene, capturing how the various objects relate to one another and translating it all into natural-sounding language. Many efforts to construct computer-generated natural descriptions of images propose combining current state-of-the-art techniques in both computer vision and natural language processing to form a complete image description approach. But what if we instead merged recent computer vision and language models into a single jointly trained system, taking an image and directly producing a human readable sequence of words to describe it? This idea comes from recent advances in machine translation between languages, where a Recurrent Neural Network (RNN) transforms, say, a French sentence into a vector representation, and a second RNN uses that vector representation to generate a target sentence in German. Now, what if we replaced that first RNN and its input words with a deep Convolutional Neural Network (CNN) trained to classify objects in images? Normally, the CNN's last layer is used in a final Softmax among known classes of objects, assigning a probability that each object might be in the image. But if we remove that final layer, we can instead feed the CNN's rich encoding of the image into a RNN designed to produce phrases. We can then train the whole system directly on images and their captions, so it maximizes the likelihood that descriptions it produces best match the training descriptions for each image.
Our experiments with this system on several openly published datasets, including Pascal, Flickr8k, Flickr30k and SBU, show how robust the qualitative results are -- the generated sentences are quite reasonable. It also performs well in quantitative evaluations with the Bilingual Evaluation Understudy (BLEU), a metric used in machine translation to evaluate the quality of generated sentences. A picture may be worth a thousand words, but sometimes it's the words that are most useful -- so it's important we figure out ways to translate from images to words automatically and accurately. As the datasets suited to learning image descriptions grow and mature, so will the performance of end-to-end approaches like this. We look forward to continuing developments in systems that can read images and generate good natural-language descriptions. To get more details about the framework used to generate descriptions from images, as well as the model evaluation, read the full paper here." Vgl. auch: https://news.ycombinator.com/item?id=8621658.

Source

http://googleresearch.blogspot.de/2014/11/a-picture-is-worth-thousand-coherent.html

0.0029000505 = product of:
  0.005800101 = sum of:
    0.005800101 = product of:
      0.011600202 = sum of:
        0.011600202 = weight(_text_:a in 5236) [ClassicSimilarity], result of:
          0.011600202 = score(doc=5236,freq=12.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.21843673 = fieldWeight in 5236, product of:
              3.4641016 = tf(freq=12.0), with freq of:
                12.0 = termFreq=12.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.0546875 = fieldNorm(doc=5236)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Abstract

The Black Book Interactive Project at the University of Kansas (KU) is developing an expanded corpus of novels by African American authors, with an emphasis on lesser known writers and a goal of expanding research in this field. Using a custom metadata schema with an emphasis on race-related elements, each novel is analyzed for a variety of elements such as literary style, targeted content analysis, historical context, and other areas. Librarians at KU have worked to develop a variety of computational text analysis processes designed to assist with specific aspects of this metadata collection, including text mining and natural language processing, automated subject extraction based on word sense disambiguation, harvesting data from Wikidata, and other actions.

Type

a

0.0028703054 = product of:
  0.005740611 = sum of:
    0.005740611 = product of:
      0.011481222 = sum of:
        0.011481222 = weight(_text_:a in 1868) [ClassicSimilarity], result of:
          0.011481222 = score(doc=1868,freq=16.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.2161963 = fieldWeight in 1868, product of:
              4.0 = tf(freq=16.0), with freq of:
                16.0 = termFreq=16.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046875 = fieldNorm(doc=1868)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

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.

Type

a

0.0026473717 = product of:
  0.0052947435 = sum of:
    0.0052947435 = product of:
      0.010589487 = sum of:
        0.010589487 = weight(_text_:a in 2933) [ClassicSimilarity], result of:
          0.010589487 = score(doc=2933,freq=10.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.19940455 = fieldWeight in 2933, product of:
              3.1622777 = tf(freq=10.0), with freq of:
                10.0 = termFreq=10.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.0546875 = fieldNorm(doc=2933)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Abstract

We propose a method for improving access to scientific literature by analyzing the content of research papers beyond citation links and topic tracking. Our model relies on a typology of explicit semantic relations. These relations are instantiated in the abstract/introduction part of the papers and can be identified automatically using textual data and external ontologies. Preliminary results show a promising precision in unsupervised relationship classification.

Type

a

0.0024857575 = product of:
  0.004971515 = sum of:
    0.004971515 = product of:
      0.00994303 = sum of:
        0.00994303 = weight(_text_:a in 1167) [ClassicSimilarity], result of:
          0.00994303 = score(doc=1167,freq=12.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.18723148 = fieldWeight in 1167, product of:
              3.4641016 = tf(freq=12.0), with freq of:
                12.0 = termFreq=12.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046875 = fieldNorm(doc=1167)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Abstract

The Indiana University School of Library and Information Science opened a new research laboratory in January 2003; The Indiana University School of Library and Information Science Information Processing Laboratory [IU IP Lab]. The purpose of the new laboratory is to facilitate collaboration between scientists in the department in the areas of information retrieval (IR) and information visualization (IV) research. The lab has several areas of focus. These include grid and cluster computing, and a standard Java-based software platform to support plug and play research datasets, a selection of standard IR modules and standard IV algorithms. Future development includes software to enable researchers to contribute datasets, IR algorithms, and visualization algorithms into the standard environment. We decided early on to use OAI-PMH as a resource discovery tool because it is consistent with our mission.

Type

a

0.0024857575 = product of:
  0.004971515 = sum of:
    0.004971515 = product of:
      0.00994303 = sum of:
        0.00994303 = weight(_text_:a in 1557) [ClassicSimilarity], result of:
          0.00994303 = score(doc=1557,freq=12.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.18723148 = fieldWeight in 1557, product of:
              3.4641016 = tf(freq=12.0), with freq of:
                12.0 = termFreq=12.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046875 = fieldNorm(doc=1557)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

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.

Type

a

0.0024857575 = product of:
  0.004971515 = sum of:
    0.004971515 = product of:
      0.00994303 = sum of:
        0.00994303 = weight(_text_:a in 1871) [ClassicSimilarity], result of:
          0.00994303 = score(doc=1871,freq=12.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.18723148 = fieldWeight in 1871, product of:
              3.4641016 = tf(freq=12.0), with freq of:
                12.0 = termFreq=12.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046875 = fieldNorm(doc=1871)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Abstract

Inspired by recent advances in multimodal learning and machine translation, we introduce an encoder-decoder pipeline that learns (a): a multimodal joint embedding space with images and text and (b): a novel language model for decoding distributed representations from our space. Our pipeline effectively unifies joint image-text embedding models with multimodal neural language models. We introduce the structure-content neural language model that disentangles the structure of a sentence to its content, conditioned on representations produced by the encoder. The encoder allows one to rank images and sentences while the decoder can generate novel descriptions from scratch. Using LSTM to encode sentences, we match the state-of-the-art performance on Flickr8K and Flickr30K without using object detections. We also set new best results when using the 19-layer Oxford convolutional network. Furthermore we show that with linear encoders, the learned embedding space captures multimodal regularities in terms of vector space arithmetic e.g. *image of a blue car* - "blue" + "red" is near images of red cars. Sample captions generated for 800 images are made available for comparison.

Type

a

0.0023678814 = product of:
  0.0047357627 = sum of:
    0.0047357627 = product of:
      0.009471525 = sum of:
        0.009471525 = weight(_text_:a in 2930) [ClassicSimilarity], result of:
          0.009471525 = score(doc=2930,freq=8.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.17835285 = fieldWeight in 2930, product of:
              2.828427 = tf(freq=8.0), with freq of:
                8.0 = termFreq=8.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.0546875 = fieldNorm(doc=2930)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Abstract

We present a framework for keyphrase extraction from scientific journals in diverse research fields. While journal articles are often provided with manually assigned keywords, it is not clear how to automatically extract keywords and measure their significance for a set of journal articles. We compare extracted keyphrases from journals in the fields of astrophysics, mathematics, physics, and computer science. We show that the presented statistics-based framework is able to demonstrate differences among journals, and that the extracted keyphrases can be used to represent journal or conference research topics, dynamics, and specificity.

Type

a

0.002269176 = product of:
  0.004538352 = sum of:
    0.004538352 = product of:
      0.009076704 = sum of:
        0.009076704 = weight(_text_:a in 5212) [ClassicSimilarity], result of:
          0.009076704 = score(doc=5212,freq=10.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.1709182 = fieldWeight in 5212, product of:
              3.1622777 = tf(freq=10.0), with freq of:
                10.0 = termFreq=10.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046875 = fieldNorm(doc=5212)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Abstract

"Alle paar Jahre wird die Bibliothekscommunity mit Veränderungen in der inhaltlichen Erschließung durch die Deutsche Nationalbibliothek konfrontiert. Sicher werden sich viele noch an die Einschnitte des Jahres 2014 für die Reihe A erinnern: Seither werden u.a. Ratgeber, Sprachwörterbücher, Reiseführer und Kochbücher nicht mehr mit Schlagwörtern erschlossen (vgl. das DNB-Konzept von 2014). Das Jahr 2017 brachte die Einführung der maschinellen Indexierung für die Reihen B und H bei gleichzeitigem Verlust der DDC-Tiefenerschließung (vgl. DNB-Informationen von 2017). Virulent war seither die Frage, was mit der Reihe A passieren würde. Seit wenigen Tagen kann man dies nun auf der Website der DNB nachlesen. (Nebenbei: Es ist zu befürchten, dass viele Links in diesem Blog-Beitrag in absehbarer Zeit nicht mehr funktionieren werden, da ein Relaunch der DNB-Website angekündigt ist. Wie beim letzten Mal wird es vermutlich auch diesmal keine Weiterleitungen von den alten auf die neuen URLs geben.)"

Source

https://www.basiswissen-rda.de/dnb-sacherschliessung-reihen-a-und-b/

Type

a

0.0022374375 = product of:
  0.004474875 = sum of:
    0.004474875 = product of:
      0.00894975 = sum of:
        0.00894975 = weight(_text_:a in 1869) [ClassicSimilarity], result of:
          0.00894975 = score(doc=1869,freq=14.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.1685276 = fieldWeight in 1869, product of:
              3.7416575 = tf(freq=14.0), with freq of:
                14.0 = termFreq=14.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.0390625 = fieldNorm(doc=1869)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Abstract

Automatically describing the content of an image is a fundamental problem in artificial intelligence that connects computer vision and natural language processing. In this paper, we present a generative model based on a deep recurrent architecture that combines recent advances in computer vision and machine translation and that can be used to generate natural sentences describing an image. The model is trained to maximize the likelihood of the target description sentence given the training image. Experiments on several datasets show the accuracy of the model and the fluency of the language it learns solely from image descriptions. Our model is often quite accurate, which we verify both qualitatively and quantitatively. For instance, while the current state-of-the-art BLEU-1 score (the higher the better) on the Pascal dataset is 25, our approach yields 59, to be compared to human performance around 69. We also show BLEU-1 score improvements on Flickr30k, from 56 to 66, and on SBU, from 19 to 28. Lastly, on the newly released COCO dataset, we achieve a BLEU-4 of 27.7, which is the current state-of-the-art.

Content

Vgl.: http://googleresearch.blogspot.de/2014/11/a-picture-is-worth-thousand-coherent.html. Vgl. auch: https://news.ycombinator.com/item?id=8621658.

0.0020714647 = product of:
  0.0041429293 = sum of:
    0.0041429293 = product of:
      0.008285859 = sum of:
        0.008285859 = weight(_text_:a in 4309) [ClassicSimilarity], result of:
          0.008285859 = score(doc=4309,freq=12.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.15602624 = fieldWeight in 4309, product of:
              3.4641016 = tf(freq=12.0), with freq of:
                12.0 = termFreq=12.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.0390625 = fieldNorm(doc=4309)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Abstract

Semantic annotations have to satisfy quality constraints to be useful for digital libraries, which is particularly challenging on large and diverse datasets. Confidence scores of multi-label classification methods typically refer only to the relevance of particular subjects, disregarding indicators of insufficient content representation at the document-level. Therefore, we propose a novel approach that detects documents rather than concepts where quality criteria are met. Our approach uses a deep, multi-layered regression architecture, which comprises a variety of content-based indicators. We evaluated multiple configurations using text collections from law and economics, where the available content is restricted to very short texts. Notably, we demonstrate that the proposed quality estimation technique can determine subsets of the previously unseen data where considerable gains in document-level recall can be achieved, while upholding precision at the same time. Hence, the approach effectively performs a filtering that ensures high data quality standards in operative information retrieval systems.

Content

This is an authors' manuscript version of a paper accepted for proceedings of TPDL-2018, Porto, Portugal, Sept 10-13. The nal authenticated publication is available online at https://doi.org/will be added as soon as available.

Type

a

0.0020296127 = product of:
  0.0040592253 = sum of:
    0.0040592253 = product of:
      0.008118451 = sum of:
        0.008118451 = weight(_text_:a in 2144) [ClassicSimilarity], result of:
          0.008118451 = score(doc=2144,freq=8.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.15287387 = fieldWeight in 2144, product of:
              2.828427 = tf(freq=8.0), with freq of:
                8.0 = termFreq=8.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046875 = fieldNorm(doc=2144)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Abstract

The complexity and diversity of archival resources make constructing rich metadata records time consuming and expensive, which in turn limits access to these valuable materials. However, significant automation of the metadata creation process would dramatically reduce the cost of providing access points, improve access to individual resources, and establish connections between resources that would otherwise remain unknown. Using a case study at Oregon Health & Science University as a lens to examine the conceptual and technical challenges associated with automated extraction of access points, we discuss using publically accessible API's to extract entities (i.e. people, places, concepts, etc.) from digital and digitized objects. We describe why Linked Open Data is not well suited for a use case such as ours. We conclude with recommendations about how this method can be used in archives as well as for other library applications.

Type

a

0.001913537 = product of:
  0.003827074 = sum of:
    0.003827074 = product of:
      0.007654148 = sum of:
        0.007654148 = weight(_text_:a in 466) [ClassicSimilarity], result of:
          0.007654148 = score(doc=466,freq=4.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.14413087 = fieldWeight in 466, product of:
              2.0 = tf(freq=4.0), with freq of:
                4.0 = termFreq=4.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.0625 = fieldNorm(doc=466)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Abstract

We present an approach for detecting multiword phrases in mathematical text corpora. The method used is based on characteristic features of mathematical terminology. It makes use of a software tool named Lingo which allows to identify words by means of previously defined dictionaries for specific word classes as adjectives, personal names or nouns. The detection of multiword groups is done algorithmically. Possible advantages of the method for indexing and information retrieval and conclusions for applying dictionary-based methods of automatic indexing instead of stemming procedures are discussed.

Type

a

0.001843099 = product of:
  0.003686198 = sum of:
    0.003686198 = product of:
      0.007372396 = sum of:
        0.007372396 = weight(_text_:a in 1875) [ClassicSimilarity], result of:
          0.007372396 = score(doc=1875,freq=38.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.13882536 = fieldWeight in 1875, product of:
              6.164414 = tf(freq=38.0), with freq of:
                38.0 = termFreq=38.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.01953125 = fieldNorm(doc=1875)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Content

"Until now, so-called computer vision has largely been limited to recognizing individual objects. The new software, described on Monday by researchers at Google and at Stanford University, teaches itself to identify entire scenes: a group of young men playing Frisbee, for example, or a herd of elephants marching on a grassy plain. The software then writes a caption in English describing the picture. Compared with human observations, the researchers found, the computer-written descriptions are surprisingly accurate. The advances may make it possible to better catalog and search for the billions of images and hours of video available online, which are often poorly described and archived. At the moment, search engines like Google rely largely on written language accompanying an image or video to ascertain what it contains. "I consider the pixel data in images and video to be the dark matter of the Internet," said Fei-Fei Li, director of the Stanford Artificial Intelligence Laboratory, who led the research with Andrej Karpathy, a graduate student. "We are now starting to illuminate it." Dr. Li and Mr. Karpathy published their research as a Stanford University technical report. The Google team published their paper on arXiv.org, an open source site hosted by Cornell University.
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."
Computer vision specialists said that despite the improvements, these software systems had made only limited progress toward the goal of digitally duplicating human vision and, even more elusive, understanding. "I don't know that I would say this is 'understanding' in the sense we want," said John R. Smith, a senior manager at I.B.M.'s T.J. Watson Research Center in Yorktown Heights, N.Y. "I think even the ability to generate language here is very limited." But the Google and Stanford teams said that they expect to see significant increases in accuracy as they improve their software and train these programs with larger sets of annotated images. A research group led by Tamara L. Berg, a computer scientist at the University of North Carolina at Chapel Hill, is training a neural network with one million images annotated by humans. "You're trying to tell the story behind the image," she said. "A natural scene will be very complex, and you want to pick out the most important objects in the image.""

Footnote

A version of this article appears in print on November 18, 2014, on page A13 of the New York edition with the headline: Advance Reported in Content-Recognition Software. Vgl.: http://cs.stanford.edu/people/karpathy/cvpr2015.pdf. Vgl. auch: http://googleresearch.blogspot.de/2014/11/a-picture-is-worth-thousand-coherent.html. https://news.ycombinator.com/item?id=8621658 Vgl. auch: https://news.ycombinator.com/item?id=8621658.

0.001674345 = product of:
  0.00334869 = sum of:
    0.00334869 = product of:
      0.00669738 = sum of:
        0.00669738 = weight(_text_:a in 1717) [ClassicSimilarity], result of:
          0.00669738 = score(doc=1717,freq=4.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.12611452 = fieldWeight in 1717, product of:
              2.0 = tf(freq=4.0), with freq of:
                4.0 = termFreq=4.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.0546875 = fieldNorm(doc=1717)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Abstract

The German subject headings authority file (Schlagwortnormdatei/SWD) provides a broad controlled vocabulary for indexing documents of all subjects. Traditionally used for intellectual subject cataloguing primarily of books the Deutsche Nationalbibliothek (DNB, German National Library) has been working on developping and implementing procedures for automated assignment of subject headings for online publications. This project, its results and problems are sketched in the paper.

Type

a

0.0014647468 = product of:
  0.0029294936 = sum of:
    0.0029294936 = product of:
      0.005858987 = sum of:
        0.005858987 = weight(_text_:a in 1873) [ClassicSimilarity], result of:
          0.005858987 = score(doc=1873,freq=6.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.11032722 = fieldWeight in 1873, product of:
              2.4494898 = tf(freq=6.0), with freq of:
                6.0 = termFreq=6.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.0390625 = fieldNorm(doc=1873)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

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.

Type

a

0.0014647468 = product of:
  0.0029294936 = sum of:
    0.0029294936 = product of:
      0.005858987 = sum of:
        0.005858987 = weight(_text_:a in 3105) [ClassicSimilarity], result of:
          0.005858987 = score(doc=3105,freq=6.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.11032722 = fieldWeight in 3105, product of:
              2.4494898 = tf(freq=6.0), with freq of:
                6.0 = termFreq=6.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.0390625 = fieldNorm(doc=3105)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Type

a

0.0014351527 = product of:
  0.0028703054 = sum of:
    0.0028703054 = product of:
      0.005740611 = sum of:
        0.005740611 = weight(_text_:a in 5851) [ClassicSimilarity], result of:
          0.005740611 = score(doc=5851,freq=4.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.10809815 = fieldWeight in 5851, product of:
              2.0 = tf(freq=4.0), with freq of:
                4.0 = termFreq=4.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046875 = fieldNorm(doc=5851)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Type

a

0.001353075 = product of:
  0.00270615 = sum of:
    0.00270615 = product of:
      0.0054123 = sum of:
        0.0054123 = weight(_text_:a in 2596) [ClassicSimilarity], result of:
          0.0054123 = score(doc=2596,freq=8.0), product of:
            0.053105544 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.046056706 = queryNorm
            0.10191591 = fieldWeight in 2596, product of:
              2.828427 = tf(freq=8.0), with freq of:
                8.0 = termFreq=8.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.03125 = fieldNorm(doc=2596)
      0.5 = coord(1/2)
  0.5 = coord(1/2)

Content

Ramana Rao (Inxight, Palo Alto, CA) 7 ± 2 Insights on achieving Effective Information Access Session One: Updates and a twelve month perspective Danny Sullivan (Search Engine Watch, US / England) Portalization and other search trends Carol Tenopir (University of Tennessee) Search realities faced by end users and professional searchers Session Two: Today's search engines and beyond Daniel Hoogterp (Retrieval Technologies, McLean, VA) Effective presentation and utilization of search techniques Rick Kenny (Fulcrum Technologies, Ontario, Canada) Beyond document clustering: The knowledge impact statement Gary Stock (Ingenius, Kalamazoo, MI) Automated change monitoring Gary Culliss (Direct Hit, Wellesley Hills, MA) User popularity ranked search engines Byron Dom (IBM, CA) Automatically finding the best pages on the World Wide Web (CLEVER) Peter Tomassi (LookSmart, San Francisco, CA) Adding human intellect to search technology Session Three: Panel discussion: Human v automated categorization and editing Ev Brenner (New York, NY)- Chairman James Callan (University of Massachusetts, MA) Marc Krellenstein (Northern Light Technology, Cambridge, MA) Dan Miller (Ask Jeeves, Berkeley, CA) Session Four: Updates and a twelve month perspective Steve Arnold (AIT, Harrods Creek, KY) Review: The leading edge in search and retrieval software Ellen Voorhees (NIST, Gaithersburg, MD) TREC update Session Five: Search engines now and beyond Intelligent Agents John Snyder (Muscat, Cambridge, England) Practical issues behind intelligent agents Text summarization Therese Firmin, (Dept of Defense, Ft George G. Meade, MD) The TIPSTER/SUMMAC evaluation of automatic text summarization systems Cross language searching Elizabeth Liddy (TextWise, Syracuse, NY) A conceptual interlingua approach to cross-language retrieval. Video search and retrieval Armon Amir (IBM, Almaden, CA) CueVideo: Modular system for automatic indexing and browsing of video/audio Speech recognition Michael Witbrock (Lycos, Waltham, MA) Retrieval of spoken documents Visualization James A. Wise (Integral Visuals, Richland, WA) Information visualization in the new millennium: Emerging science or passing fashion? Text mining David Evans (Claritech, Pittsburgh, PA) Text mining - towards decision support