The present invention relates to data processing and more particularly, but not exclusively, relates to the visualization of data from a data stream.
Recent technological advancements have led to the generation of vast amounts of electronic data. Unfortunately, the ability to quickly identify patterns or relationships, and/or the ability to readily perceive underlying concepts from such data remain highly limited. Moreover, the ability to meaningfully visualize large amounts of data presents an even greater challenge.
Data of interest can be static in nature, such as that stored in a given corpus; or dynamic in nature, such as a data stream. As a more recent area of interest, data stream processing poses several unique challenges. The time-varying information of a data stream typically has the characteristics of arriving continuously, unpredictably, and unboundedly without any persistent patterns. Data stream examples include newswires, internet click streams, network resource measurements, phone call records, and remote sensing imagery, just to name a few. The increasing desire to more immediately analyze data streams in various time-sensitive applications (such as Homeland Security) has created a particular interest in better visualization tools for data streams.
Thus, there is an ongoing need for further contributions in this area of technology.
One embodiment of the present invention is a unique data processing technique. Other embodiments include unique apparatus, systems, devices, and methods for visualizing data stream information.
A further embodiment of the present invention includes a technique to analyze a data stream, including: processing a stream of data vectors; generating a visualization from a first group of these vectors; determining a set of values corresponding to one or more eigenvectors for a matrix defined with this first group of data vectors; and projecting each member of a second group of data vectors onto the visualization as a function of the set of values.
Yet a further embodiment includes: receiving a first portion of a data stream at or below a defined rate; generating a first visualization from a group of data vectors corresponding to the first portion; receiving a second portion of the data stream above the defined rate; and generating a second visualization by updating the first visualization with one or more additional data vectors as a function of an eigenspace defined within the group of data vectors. The one or more additional vectors correspond to the second portion of the data stream. In one form, the generation of the second visualization includes determining a dot product between each of the one or more additional data vectors and one or more eigenvectors corresponding to the eigenspace.
Still another embodiment of the present invention comprises: receiving a data stream, processing a group of data vectors corresponding to the data stream, generating a reduced data set which includes reducing dimension of the data vectors as a function of wavelet decomposition, and providing a representation with a reduced data set corresponding to a visualization of a portion of the data stream. In one form, a wavelet decomposition routine is performed with Haar wavelets. Alternatively or additionally, a multidimensional scaling routine is performed to provide a visualization.
Another embodiment of the present invention includes: receiving a data stream, processing a stream of data vectors corresponding to the data streams, visualizing at least a portion of the data stream by executing a multidimensional scaling routine with at least a corresponding portion of the data vectors, and performing a routine with a group of data vectors to provide a data set with a reduced number of data elements relative to the number of elements in the group. This routine can include dimension reduction and/or sampling of the group of data vectors.
In further embodiments, methods, systems, apparatus, and devices are provided to perform the various above embodiments. In one example, a device carrying logic executable by a processing subsystem receives a data stream to perform any of the various embodiments indicated above. In another example, a system is provided with a data communication subsystem, a processing subsystem, and an operator input/output subsystem arranged to perform any of the above embodiments.
Accordingly, one object of the present invention is to provide a unique data processing technique.
Another object is to provide a unique apparatus, system, device, or method for the visualization of data stream information.
Further objects, embodiments, forms, features, aspects, benefits, and advantages of the present invention will become apparent from the drawings and detailed description contained herein.
For the purpose of promoting an understanding of the principles of the invention, reference will now be made to the embodiments illustrated in the drawings and specific language will be used to describe the same. It will nevertheless be understood that no limitation of the scope of the invention is thereby intended. Any alterations and further modifications in the described embodiments, and any further applications of the principles of the invention as described herein are contemplated as would normally occur to one skilled in the art to which the invention relates.
Among the embodiments of the present invention are various techniques to dynamically visualize transient data streams. In at least some of these embodiments there is an adaptive/selective application of Multidimensional Scaling (MDS). While the time-sensitive nature of these data streams often results in the desire to be immediately responsive to changes in many applications, the unpredictable and unbounded characteristics of this information can potentially overwhelm many MDS algorithms that require a full re-computation for every update. In one form, an adaptive visualization technique based on data stratification ingests stream information adaptively when influx rate exceeds processing rate. In another form that may be alternatively or additionally employed, an incremental visualization technique based on data fusion projects new data stream information directly onto an orthogonal visualization subspace spanned by selected singular vectors of the previously processed neighboring data. In this approach, a sliding pair of data windows can be utilized in which the dominant eigenvectors obtained from the larger of the windows define this subspace and are used to map the information from the smaller of the windows onto a visualization without re-processing the entire dataset.
In one implementation of the present invention, the visualization output of an MDS analysis is a low-dimensional scatter plot in which pairwise distances between any points reflect the similarities of the items represented by the points. To assess accuracy of the progressive/adaptive visualization techniques used to represent a changing data stream, error-tracking can be utilized based on visual and/or computational comparative analysis of resulting scatter plots.
Operator I/O subsystem 21b includes operator input devices 24 and operator output devices 26 operatively coupled to processing subsystem 21a. Input devices 24 include a conventional mouse 24a and keyboard 24b, and alternatively or additionally can include a trackball, light pen, voice recognition subsystem, and/or different input device type as would occur to those skilled in the art. Output devices 26 include a conventional graphic display 26a, such as a color or noncolor plasma, Cathode Ray Tube (CRT), or Liquid Crystal Display (LCD) type, and color or noncolor printer 26b. Alternatively or additionally output devices 26 can include an aural output system and/or different output device type as would occur to those skilled in the art. Further, in other embodiments, more or fewer operator input devices 24 or operator output devices 26 may be utilized.
System 20 also includes computer network 30 coupled to computer 21 via data communication subsystem 21c. Network 30 can be a Local Area Network (LAN); Municipal Area Network (MAN); Wide Area Network (WAN), such as the Internet; another type as would occur to those skilled in the art; or a combination of these. One or more corpora 32 are coupled to computer equipment 21 via network 30. Computer equipment 21 is operable to selectively process data selected from one or more corpora 32. Network 30 also couples computer 40 to computer equipment 21; where computer 40 is remotely located relative to computer equipment 21. Computer 40 can include a processor, input devices, output devices, and/or memory as described in connection with computer equipment 21; however, these features of computer 40 are not shown to preserve clarity.
Computer 40 and computer equipment 21 can be arranged as client and server, respectively, in relation to some or all of the data processing of the present invention. For this arrangement, it should be understood that many other remote computers 40 could be included as clients of computer equipment 21, but are not shown to preserve clarity. In another embodiment, computer equipment 21 and computer 40 can both be participating members of a distributed processing arrangement with one or more processors located at a different site relative to the others. The distributed processors of such an arrangement can be used collectively to execute processes, routines, operations, stages, conditionals, and/or procedures according to the present invention. In still other embodiments, remote computer 40 may be absent.
In addition to computer network 30, data communication subsystem 21c is also operatively coupled to data stream sources 50. Sources 50 are also indirectly connected to subsystem 21c via computer network 30. Sources 50 include newswire source 52 which provides a data stream if textual documents regarding pertinent news stories, and image source 54 which provides streaming image data. Sources 50 also include one or more sensors 56 providing one or more data streams. Still other data stream sources are designated by reference numeral 58, which can include audio data, video data, phone call records, internet click streams, and/or travel records, just to name a few examples. It should be appreciated that data communicated to computer equipment 21 from network 30 can be presented as a data stream whether originating from the one or more corpora 32, computer 40, sources 50, and/or elsewhere.
Operating logic for processing subsystem 21a is arranged to facilitate performance of various routines, procedures, stages, operations, and/or conditionals described hereinafter in connection with
From operation 122, process 120 continues with conditional 124. Conditional 124 tests whether the format of data stream DS is acceptable for visualization processing directly or whether any reformatting/preprocessing is required. In subsequent operations of process 120, information from data stream DS is input as one or more corresponding data vectors. Data stream DS may be received in an appropriate data vector format, in which case the test of conditional 124 is affirmative (true), and process 120 continues with visualization operation 130 to be more fully described hereinafter. In the case where the format of data stream DS is not acceptable, such that the test of conditional 124 is negative (false), then process 120 continues with operation 126. In operation 126, data stream DS is formatted to provide the desired data vector arrangement. The specific nature of operation 126 varies with the type of information provided by data stream DS and its organization. In one example, pixelated image data may be received in a vectorized form requiring little (if any) reformatting in operation 126. Such a data stream DS could be provided from image source 54.
In another example, a textual document form of data stream DS is received that requires more extensive reformatting/preprocessing in operation 126. Typically, a document data stream has one or more aspects in common with respect to the different documents represented, such as document type, overall topic, and the like; however, documents from diverse collections/sources can alternatively be utilized. In one particular example further described hereinafter, a document type of data stream DS is directed to news stories of the variety that could be provided by newswire source 52.
For a document data stream of this type, operation 126 includes a term standardization routine in which a set of terms S is determined for later processing. Such standardization can include typical stemming, identification of phrases (i.e., word sequences that should be treated as one unit), and mapping known synonyms to a common canonical form. Typically, functional words or ‘stop’ words will be removed when determining this standardized lexicon. Functional words include modifiers such as ‘a’, ‘the’, and ‘this’ that are necessary for grammatical comprehension but do not directly contribute to a concept. Functional words can be removed by comparing them with a list of known functional terms—a ‘stop-word’ list. Alternatively, if a stop-word list is not available (for example, if a foreign language is being analyzed for which a stop-word list is not known), functional words can be identified automatically via a topicality calculation executed with system 20.
In such a calculation for a given term, let A be the number of documents that contain the term. Let N be the number of documents in the test collection, and let T be the total number of times the term occurs in the collection. Then if the term is distributed randomly T times across the N documents, it would be expected to occur in
documents. If the term occurs in significantly more documents than expected by chance, it is considered to be regularly distributed, typical of a functional word. Thus, functional terms can be automatically identified as those terms for which
where λ is a threshold that may have been selected based on previous experience, or based on statistical considerations. In one embodiment, λ=0.25 has been found to be adequate for English documents. A. Bookstein, S. T. Klein, and T. Raita, “Clumping Properties of Content-Bearing Words” Journal of the American Society for Information Science (published on the world wide web 1998) is cited as a source of background information concerning such approaches.
Further, for a text document type of data stream DS, operation 126 can generate a document feature space as a function of the term set S. In one form, the resulting data vectors each represent a different document and are each organized to contain an array of real numbers. These real numbers each represent a weighted strength of one of the terms S. A sequence of vectors corresponding to a quantity desired to present an initial visualization is normalized to provide a corresponding document matrix.
In one newswire example, a corpus was used consisting of 3,298 news articles collected from open sources during the date range of Apr. 20-26, 1995. This corpus has a strong theme associated with the bombing of the U.S. Federal Building in Oklahoma, the O. J. Simpson trial, and the French elections. The corpus was processed to identify a set of content-bearing words from the documents. Words separated by white spaces in a corpus are evaluated within the context of the corpus to assess whether a word is interesting enough to be a topic. The co-occurrence or lack of co-occurrence of these words in the documents was used to evaluate the strengths of the words. From this set of words, document vectors were constructed for the newswire corpus. In this instance, a document vector, which was an array of real numbers, contained the weighted strengths of the interesting words found in the corresponding document. These vectors were normalized and the resulting document matrix represented the corpus. For this example, a document vector dimension of 200 numbers (elements) was utilized. Because there are 3,298 documents in the newswire corpus, the dimensions of the document matrix are 3,298×200. Correspondingly, 200×3298=659,600 data elements are in this matrix.
Document vectors can be provided in the form of a term-by-document frequency matrix; where, the (ij)th entry contains the frequency of the ith term in the jth document, an example of which follows in Table 1:
It should be understood that in other embodiments, a term-by-document frequency matrix can include fewer, but typically, many more documents and/or terms. Alternatively or additionally, the frequency can be weighted based on one or more criteria, such as an information-theoretic measure of content or information contained in a given term and/or document. In one such form, term frequencies are weighted by a measure of their content relative to their prevalence in the document collection. To standardize for documents of varying sizes, the columns of a weighted term-by-document frequency matrix might also be normalized prior to analysis.
A term-by-document frequency matrix is often useful in discovering co-occurrence patterns of terms, which can often correspond to underlying concepts. First-order co-occurrence patterns relate terms that frequently occur together in the same documents; second-order co-occurrence patterns relate terms that have similar first-order co-occurrence patterns, so that two terms can be related by second-order co-occurrence even if they never occur together in a document. As previously described, co-occurrence aspects can be used in determining the strength of words for potential inclusion in the document data vectors.
From operation 126, process 120 continues with operation 130. Alternatively, operation 130 is reached directly, bypassing operation 126, if the test of conditional 124 is affirmative (true). In operation 130, a baseline visualization is generated from the document matrix received from the preprocessing of operation 126 or directly from data stream DS. The visualization can be completely or partially displayed with one or more of output devices 26. In one form, the visualization is created by reducing the relatively high-dimension document matrix into a visualization of fewer dimensions using Multidimensional Scaling (MDS). MDS includes a variety of scaling techniques subject to categorization as classical or nonclassical, metric or nonmetric, and the like. Adhering to the terminology in the book: Multidimensional Scaling by Cox, T. F. and Cox, M. A. A. (1994, Chapman & Hall); both the classical MDS approach and the least-squares MDS approach (alternatively designated a Sammon projection) were used to experimentally demonstrate various aspects of the present application. Beside this reference, other sources of various MDS algorithms, procedures, and/or routines include International Patent Application Publication No. WO 00/39705, published 6 Jul. 2000; Buga & Swaywe et al., “XGvis: Interactive Data Visualization with Multidimensional Scaling” (Published 29 Nov. 2001); “Evaluating Similarity-based Visualizations as Interfaces for Image Browsing”, Technical Report No. 543, University of Cambridge Computer Laboratory, (published September 2002); and U.S. Pat. No. 6,615,211 B2 to Beygelzimer et al. It should be understood, that in other embodiments of the present invention, different scaling techniques could be alternatively or additionally utilized. As an option, clustering (such as K-mean clustering) and/or various other techniques can be utilized to modify the MDS visualization as displayed. Further information regarding such clustering techniques and corresponding algorithms may be found in Seber, G. A. F., Multivariate Observations, (1984, John Wiley & Sons).
Given a high-dimensional dataset (such as a set of similar data objects represented by numerical vectors), MDS generates a low-dimensional configuration—like a 2-D scatterplot—such that the pairwise distances between any points in the low-dimensional space approximate the similarities between the vectors that represent the points. In one nonlimiting example,
In response to a degree of change in the data stream DS influx or receive rate relative to the processing rate of system 20, adaptive visualization according to one embodiment of the present invention is based on data stratification that substantially reduces the processing time of the data stream DS and yet generally maintains the overall integrity of the visualization output. In one particular version, if the primary data processing route has overflowed, the data from data stream DS is re-directed to a secondary route. This second route generates a coarser version of visualization but at a much faster processing rate. And if the secondary route also overflows, the data from data stream DS is re-directed to a tertiary route and so on. The stratification strategies of the present invention include vector dimension reduction, vector sampling, and eigenvector data fusion.
Before describing process 120 further, these strategies are each described in more detail. Vector dimension reduction is directed to decreasing the physical size of the data vectors but maintaining most of their contents of interest. In one form, this reduction is accomplished by applying dyadic wavelets to decompose individual vectors (and thus compress them) progressively. It has been found that Haar wavelets perform well, and generally outperform all other wavelet candidates in processing time, which is generally desirable.
While the example in
Using a wavelet decomposition routine, the dimensions of the document vectors are progressively reduced from 200, to 100, and then to 50 going from left-to-right across the first row of scatter plots in
The second stratification strategy is to reduce the number of data vectors based on sampling. A regular sampling technique was used to obtain an even data distribution. Other sampling options such as a statistical-based distribution, just to name one alternative, can also be applied. With the 3,298×200 newswire document matrix previously described, color identities are assigned to each scatter point; however, instead of reducing the dimensions of the vectors, this time the number of document vectors is progressively reduced by 50% every time using a regular sampling method. In
To improve the visualization for comparison and evaluation, the two approaches are progressively combined and concatenated in the scatter plots of
Considering computational performance, Mathematica 4.2 [Mathematica 2003] running on a Macintosh G4 with 1 GB memory was utilized to report relative computation performance. Table 2 shows the results. In Table 2, the top row shows the number of dimensions in the document vectors and the left column shows the number of document vectors included in the computation. The other nine numerical entries are computation time measured in wall clock seconds. The corresponding scatter plot of each entry is shown in
The results in Table 2 show a 92.5% time reduction (from 34.9 s to 2.62 s) by compressing the vectors from 200 to 50 dimensions. The table further demonstrates a notable 97.5% time reduction (from 34.9 s to 0.89 s) by simultaneously reducing the number of vectors from 3,298 to 824 in the computation.
To show the flexibility of the adaptive visualization technique, a second scaling example using a least-squares MDS technique known as a Sammon Projection was performed. Classical MDS treats similarity between two vectors directly as Euclidean distances whereas least-squares MDS takes it as the least squares of a continuous monotonic function. With regard to visualization, one difference between a Sammon Projection and a classical MDS projection is that the former usually has fewer overlapping clusters due largely to its non-linear mapping approach.
The adaptive visualization technique can also be used to visualize other types of data streams DS. In a further experiment, efficacy with image streams was demonstrated as shown in the comparative, grayscale matrix of nine scatter plots in
One common motivation to include all spectral bands in the image analysis is that subjects appearing identical in one spectral band (like visible color) may be very different from each other if all possible spectral bands are considered. This example shows that the same stratification strategies can be favorably applied to analyze imagery streams. The scatter plots of
In addition to visual techniques of scatter plot comparison, computational techniques are also desired. In statistics studies, a class of techniques for matching two similar n-D configurations and producing a measure of the match is known as Procrustes analysis. Procrustes analysis can match scatter plots in any number of dimensions assuming the one-to-one correspondence information among the scatter points is known. Generally, given two 2-dimensional scatter plots X and Y where X and Y are (n×2) matrices, the stages to match X to Y and report a measure of the match using Procrustes analysis are as follows:
Table 3 shows the results of Procrustes analyses that were carried out on the corpus scatter plots in
Matching results for the remote sensing imagery scatter plots shown in
While vector dimension reduction and sampling strategies focus on the use of stratified vectors in place of full-resolution data sets to generate fast and accurate MDS scatter plots, such strategies do not eliminate the requirement to re-process the entire dataset whenever new items arrive. In another aspect of the present invention, a strategy to update an existing visualization without re-processing the entire dataset is provided. As previously noted, the visualization subspaces spanned by the two dominant Eigenvectors are resilient to changes. Furthermore, considering this characteristic, the hyperspectral imagery dataset is used to evaluate the similarity between the eigenvectors (and the corresponding scatter plots) generated from local image regions versus the entire dataset. To provide identities to individual pixels, the image representation shown in the upper left corner of
Specifically, a MDS scatter plot using the pixel vectors from the entire hyperspectral imagery was generated (upper left scatter plot of
Because the most dominant Eigenvector is the line though the centroid of the scatter points along which the variance of the projections is greatest (not necessarily the direction of the greatest ranges or extent of the data) and the second-most dominant Eigenvector is orthogonal to the most dominant Eigenvector, these Eigenvectors tend to be robust for changes unless a substantial amount of disparate information is added.
Using the same hyperspectral imagery to demonstrate, the top row “a” of
Notably, the scatter plots of rows a, b, and c in
Procrustes analysis can be used to evaluate the errors between a full-resolution standard scatter plot and those based on multiple sliding windows. Alternatively, to speed-up error-tracking, the full-resolution standard scatter plot can be replaced with a fast and accurate substitute like one using reduced dimension and/or sampled data vectors.
The results in Table 2 show that up to 92% of computation time (from 34.9 s to 2.62 s) can be saved by compressing the dimensions of 3,268 vectors by 75%. And the results in Table 4 show that a 75% reduced data matrix (dimension reduced from 169 to 42) can still be about as accurate as the full resolution one. Because of this faster error checking process, one can now afford to carry out error estimation more frequently and thus improve the overall quality of the analysis.
Although data vector dimension reduction, data vector sampling, eigenvector projection, and corresponding error-tracking techniques can be utilized independently, in other embodiments, one or more of these approaches can be combined. Returning to
On the other hand, if the data receive rate equals or exceeds the processing rate, an alternative strategy is utilized in the form of visualization update routine 140 as further depicted in
Returning to
If the test of conditional 132 is affirmative (true), a similarity analysis is preformed in operation 150. This operation may include comparing the baseline visualization before and after new data vector projection by Procrustes analysis and/or generating one or more overviews of the baseline visualization for comparison in accordance with routine 160, which is more fully explained hereinafter. From operation 150, conditional 134 is reached. If the test of conditional 132 is negative (false), operation 150 and conditional 134 are bypassed, and process 120 continues with conditional 138.
Conditional 134 tests whether an error threshold has been reached as determined with operation 150. If the test of conditional 134 is affirmative (true), process 120 returns to operation 130 to generate a new baseline visualization representative of data stream DS. If the test of conditional 134 is negative (false), conditional 138 is reached.
Conditional 138 tests if an overview visualization generated with one or more vector data reduction techniques is desired. If the test of conditional 138 is affirmative (true), routine 160 is executed which is shown in greater detail in the flowchart of
In
Conditional 166 tests whether a vector sampling technique is to be utilized in generating a stratified overview visualization. If the outcome of the test of conditional 166 is affirmative (true), operation 168 is performed. In operation 168, the data vectors are regularly sampled and/or sampled in accordance with a selected distribution. Operation 168 can include options/selections regarding the type of sampling and/or degree of sampling to be performed. Such sampling operation parameters can be assigned by an operator, automatically assigned, and/or may be static or dynamic in nature.
From operation 168, routine 160 continues with operation 170. Likewise, operation 170 is reached directly from conditional 166 when the outcome of conditional 166 is negative (false). Operation 170 returns a reduced data set representative of the overview visualization which can be visually presented with one or more of output devices 26, and/or computationally compared through similarity analysis in process 120.
Returning to
Indeed, many other embodiments and variations are envisioned. For example, different levels of receive rates could be used to trigger different stratification actions. In a further example, one or more of the described types of stratification are not utilized and/or a different stratification strategy is additionally or alternatively utilized instead of one or more of those included in process 120. It should further be understood that in alternative embodiments one or more of the image adjustments of process 120 could be performed with computer equipment 21 before presenting the visualization. Furthermore, stages of process 120 can be iteratively performed in response to changing a selection of one or more parameters by an operator via input device(s) 24. As such selections are made, different visualization results can be presented for comparison on display 26a and/or printer 26b of output device(s) 26. In one form, a Graphic User Interface (GUI) is provided that permits selection among a number of different visualization parameters throughout performance of process 120, such as visualization data size, coloration, resolution, etc. Further, filtering parameters and/or filter type selection (if any) can be presented as a user option and/or automatically/dynamically implemented.
GUI selection can be input by an operator with one or more of devices 24. GUI techniques can also be used to facilitate storage and presentation of visualizations in one or more forms. Indeed, it should be understood that in other embodiments, various operations, routines, and conditionals of process 120 can be combined, performed in a different order, omitted, and/or added to other processes as would occur to those skilled in the art.
Alternatively or additionally, different visualizations can be provided in different windows on display 26a that can be separately scaled, opened, or closed. Optionally, a zoom-in/zoom-out tool can be provided to rescale a visualization and/or portion of a visualization based on operator input with one or more of devices 24, and/or otherwise perform different imaging operations. In one embodiment, coloration enhancement is targeted through operator selection of specified area(s) of a visualization. In another embodiment, different image element patterns and/or appearances are used in different visualization areas. According to other embodiments of the present invention, only a portion of a computer visualization dataset may be displayed at one time and/or a may be displayed on multiple pages or windows.
Any experiments, experimental examples, or experimental results provided herein are intended to be illustrative of the present invention and should not be considered limiting or restrictive with regard to the invention scope. Further, any theory, mechanism of operation, proof, or finding stated herein is meant to further enhance understanding of the present invention and is not intended to limit the present invention in any way to such theory, mechanism of operation, proof, or finding. All publications, patents, and patent applications cited in this specification are herein incorporated by reference as if each individual publication, patent, or patent application were specifically and individually indicated to be incorporated by reference and set forth in its entirety herein. While the invention has been illustrated and described in detail in the drawings and foregoing description, the same is to be considered as illustrative and not restrictive in character, it being understood that only selected embodiments have been shown and described and that all changes, equivalents, and modifications that come within the spirit of the invention described herein or defined by the following claims are desired to be protected.
The present application is a continuation of U.S. Ser. No. 10/688,063 filed Oct. 17, 2003, which claims the benefit of U.S. Provisional App. No. 60/459,841 filed Apr. 1, 2003, which are hereby incorporated by reference.
Number | Date | Country | |
---|---|---|---|
60459841 | Apr 2003 | US |
Number | Date | Country | |
---|---|---|---|
Parent | 10688063 | Oct 2003 | US |
Child | 12436906 | US |