<html><head><meta http-equiv="Content-Type" content="text/html charset=utf-8"></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; "><div><u><b><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">Call for Papers: </span><span style="background-color: rgb(255, 255, 255); font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; text-align: left; ">Visual Analytics using Multidimensional Projections Workshop (co-located with EuroVis 2013)</span></b></u></div><div><div>Website: <a href="http://homepage.tudelft.nl/19j49/eurovis2013">http://homepage.tudelft.nl/19j49/eurovis2013</a></div></div><div><br></div><div><br></div><div><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">We solicit submissions for oral or poster presentation at the First International Workshop on Visual Analytics using Multidimensional Projections (VAMP), to be held on June 19, 2013, in Leipzig, Germany, in conjunction with the EuroVis 2013. The workshop will bring together researchers and practitioners from academia and industry to discuss the latest developments in the cross-section of information visualization, machine learning, and graph drawing. </span><span style="background-color: rgb(255, 255, 255); font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; ">We are planning a special issue for the best papers of the workshop.</span></div><div><br></div><div><br></div><div><br></div><div><b>Call for Papers</b></div><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">Dimensionality reduction is an active area in machine learning. New techniques have been proposed for more than 50 years, for instance, principal component analysis, classical scaling, isomap, probabilistic latent trait models, stochastic neighbor embedding, and neighborhood retrieval visualization. These techniques facilitate the visualization of high-dimensional data by representing data instances as points in a two-dimensional space in such a way that similar instances are modeled by nearby points and dissimilar instances are modelled by distant points.</span><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">Although many papers on these so-called “embedding” techniques are published every year, which aim to improve visual representations of high-dimensional data, it appears that these techniques have not gained popularity in the EuroVis community due to the inherent complexity of their interpretation.</span><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">At the cross-section of information visualization, machine learning, and graph drawing, this workshop will focus on issues that embedding techniques should address to bridge the gap with the information visualization community. Below is a (non-exhaustive) list of topics on which we solicit submissions for the workshop:</span><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">⎯ </span><b style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">Stability:</b><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "> Nonlinear embedding techniques are more efficient at preserving similarities than linear ones. However, non-linearities generate local optima as a result of which different initializations lead to different representations of the same data. The differences between these embeddings of the same data create confusion for the analyst, who is unable to grasp the common facts across the different visualizations. How can we design efficient and stable nonlinear embeddings?</span><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">⎯ </span><b style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">Embedding of dynamic data:</b><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "> Embedding usually projects all the data at once; when new data arrive, how can we embed these data without modifying the current embedding too much?</span><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">⎯ </span><b style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">Multiple methods:</b><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "> Each embedding algorithm necessarily comes with its own set of built-in underlying assumptions, and knowledge of these assumptions is often helpful in making sense of the visual output. How can we design black-box visualization methods that demand less understanding of underlying assumptions from the side of the analyst?</span><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">⎯ </span><b style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">Evaluation and subjectivity:</b><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "> Visual interpretation is inherently subjective. How can we help analysts to verify whether an eye-catching pattern is real/essential or whether it just happens to be an artefact?</span><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">⎯ </span><b style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">Inference and interactions:</b><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "> Nonlinear embedding techniques produce points clouds in which the axes have no meaning and pairwise distances are approximations which may have many artefacts. What kinds of analytical tasks can be performed with such embeddings? How can we better convey the meaning of the embeddings to analysts?</span><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">⎯ </span><b style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">Feedback:</b><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "> The human eye is excellent at visual analysis, and can identify regularities and anomalous data even without having to define an algorithm. How can we make use of this ability to enhance the predictive performance of machine learning and embedding techniques?</span><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">⎯ </span><b style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">Input data:</b><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "> Currently, the input data in embedding techniques typically comprises high-dimensional feature vectors or pairwise distance between objects. However, this is not always the kind of data that analysts encounter in practice. How can embeddings be constructed based on partial similarity rankings, associations or co-occurences of objects, heterogeneous data, data with missing values, relations between objects, structured objects, etc.?</span><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">⎯ </span><b style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">Optimizing embeddings for visual analysis:</b><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "> nonlinear embeddings are found by optimizing mathematical goodness-of-fit measures. Instead of using off-the-shelf embedding methods, can the measures and methods be designed so that the optimized embeddings will be good for carrying out concrete low-level or high-level analysis tasks from the visualization?</span><div><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br></span></div><div><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br></span></div><div><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><b>Submissions</b></span></div><div><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">Submissions should report new (unpublished) research results or ongoing research. Papers should be 4 pages (at most), excluding references, and 5 pages (at most), in total. Papers should be formatted in the style of EuroVis Short Papers</span><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">. Papers must be in English and must be submitted as PDF files. Papers will be published on the workshop website. </span><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">We are planning a special issue for the best papers of the workshop.</span><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">At least one author of each accepted submission will be expected to attend and present their findings at the workshop.</span><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); ">Papers should be submitted electronically no later than 23:59 UTC/GMT, Friday, March 8, 2013.</span></div><div><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br></span></div><div><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; background-color: rgb(255, 255, 255); "><br></span></div><div><b style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; ">Program committee</b></div><div><span style="font-family: 'Lucida Grande', 'Trebuchet MS', arial, Helvetica, sans-serif; line-height: 20px; text-align: left; "><div>Michael Aupetit, CEA LIST</div><div>Laurens van der Maaten, Delft University of Technology</div><div>Daniel Keim, University of Konstanz</div><div>Jean-Daniel Fekete, INRIA</div><div>Jaakko Peltonen, Aalto University</div><div>Ata Kaban, University of Birmingham</div><div>John Lee, Universite catholique de Louvain</div><div>Samuel Kaski, Aalto University</div><div>Frank-Michael Schleif, University of Bielefeld</div><div>Leishi Zhang, University of Konstanz</div><div>Michel Verleysen, Universite catholique de Louvain</div><div>Nicolas Heulot, CEA LIST</div></span></div><div><br></div></body></html>