Search (3 results, page 1 of 1)

  • × theme_ss:"Data Mining"
  • × type_ss:"el"
  • × year_i:[2000 TO 2010}
  1. Maaten, L. van den: Learning a parametric embedding by preserving local structure (2009) 0.00 
    0.0021674242 = product of:
  0.0043348484 = sum of:
    0.0043348484 = product of:
      0.008669697 = sum of:
        0.008669697 = weight(_text_:a in 3883) [ClassicSimilarity], result of:
          0.008669697 = score(doc=3883,freq=10.0), product of:
            0.043477926 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.037706986 = queryNorm
            0.19940455 = fieldWeight in 3883, 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=3883)
      0.5 = coord(1/2)
  0.5 = coord(1/2)
    Abstract
    The paper presents a new unsupervised dimensionality reduction technique, called parametric t-SNE, that learns a parametric mapping between the high-dimensional data space and the low-dimensional latent space. Parametric t-SNE learns the parametric mapping in such a way that the local structure of the data is preserved as well as possible in the latent space. We evaluate the performance of parametric t-SNE in experiments on three datasets, in which we compare it to the performance of two other unsupervised parametric dimensionality reduction techniques. The results of experiments illustrate the strong performance of parametric t-SNE, in particular, in learning settings in which the dimensionality of the latent space is relatively low.
    Type
    a
  2. Maaten, L. van den; Hinton, G.: Visualizing data using t-SNE (2008) 0.00 
    0.0019582848 = product of:
  0.0039165695 = sum of:
    0.0039165695 = product of:
      0.007833139 = sum of:
        0.007833139 = weight(_text_:a in 3888) [ClassicSimilarity], result of:
          0.007833139 = score(doc=3888,freq=16.0), product of:
            0.043477926 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.037706986 = queryNorm
            0.18016359 = fieldWeight in 3888, product of:
              4.0 = tf(freq=16.0), with freq of:
                16.0 = termFreq=16.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.0390625 = fieldNorm(doc=3888)
      0.5 = coord(1/2)
  0.5 = coord(1/2)
    Abstract
    We present a new technique called "t-SNE" that visualizes high-dimensional data by giving each datapoint a location in a two or three-dimensional map. The technique is a variation of Stochastic Neighbor Embedding (Hinton and Roweis, 2002) that is much easier to optimize, and produces significantly better visualizations by reducing the tendency to crowd points together in the center of the map. t-SNE is better than existing techniques at creating a single map that reveals structure at many different scales. This is particularly important for high-dimensional data that lie on several different, but related, low-dimensional manifolds, such as images of objects from multiple classes seen from multiple viewpoints. For visualizing the structure of very large data sets, we show how t-SNE can use random walks on neighborhood graphs to allow the implicit structure of all of the data to influence the way in which a subset of the data is displayed. We illustrate the performance of t-SNE on a wide variety of data sets and compare it with many other non-parametric visualization techniques, including Sammon mapping, Isomap, and Locally Linear Embedding. The visualizations produced by t-SNE are significantly better than those produced by the other techniques on almost all of the data sets.
    Type
    a
  3. Cohen, D.J.: From Babel to knowledge : data mining large digital collections (2006) 0.00 
    0.0013567394 = product of:
  0.0027134789 = sum of:
    0.0027134789 = product of:
      0.0054269577 = sum of:
        0.0054269577 = weight(_text_:a in 1178) [ClassicSimilarity], result of:
          0.0054269577 = score(doc=1178,freq=12.0), product of:
            0.043477926 = queryWeight, product of:
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.037706986 = queryNorm
            0.12482099 = fieldWeight in 1178, product of:
              3.4641016 = tf(freq=12.0), with freq of:
                12.0 = termFreq=12.0
              1.153047 = idf(docFreq=37942, maxDocs=44218)
              0.03125 = fieldNorm(doc=1178)
      0.5 = coord(1/2)
  0.5 = coord(1/2)
    Abstract
    In Jorge Luis Borges's curious short story The Library of Babel, the narrator describes an endless collection of books stored from floor to ceiling in a labyrinth of countless hexagonal rooms. The pages of the library's books seem to contain random sequences of letters and spaces; occasionally a few intelligible words emerge in the sea of paper and ink. Nevertheless, readers diligently, and exasperatingly, scan the shelves for coherent passages. The narrator himself has wandered numerous rooms in search of enlightenment, but with resignation he simply awaits his death and burial - which Borges explains (with signature dark humor) consists of being tossed unceremoniously over the library's banister. Borges's nightmare, of course, is a cursed vision of the research methods of disciplines such as literature, history, and philosophy, where the careful reading of books, one after the other, is supposed to lead inexorably to knowledge and understanding. Computer scientists would approach Borges's library far differently. Employing the information theory that forms the basis for search engines and other computerized techniques for assessing in one fell swoop large masses of documents, they would quickly realize the collection's incoherence though sampling and statistical methods - and wisely start looking for the library's exit. These computational methods, which allow us to find patterns, determine relationships, categorize documents, and extract information from massive corpuses, will form the basis for new tools for research in the humanities and other disciplines in the coming decade. For the past three years I have been experimenting with how to provide such end-user tools - that is, tools that harness the power of vast electronic collections while hiding much of their complicated technical plumbing. In particular, I have made extensive use of the application programming interfaces (APIs) the leading search engines provide for programmers to query their databases directly (from server to server without using their web interfaces). In addition, I have explored how one might extract information from large digital collections, from the well-curated lexicographic database WordNet to the democratic (and poorly curated) online reference work Wikipedia. While processing these digital corpuses is currently an imperfect science, even now useful tools can be created by combining various collections and methods for searching and analyzing them. And more importantly, these nascent services suggest a future in which information can be gleaned from, and sense can be made out of, even imperfect digital libraries of enormous scale. A brief examination of two approaches to data mining large digital collections hints at this future, while also providing some lessons about how to get there.
    Type
    a

Authors