Michaël Aupetit

Visualizing distortions and recovering topology in continuous projection techniques

The visualization of continuous multi-dimensional data based on their projection to a 2-dimensional space is a way to detect visually interesting patterns, as far as the projection provides a faithful image of the original data. In order to evaluate this faithfulness, we propose to visualize any measure associated to a projected datum or to a pair of projected data, by coloring the corresponding Voronoi cell in the projection space. We also define specific measures and show how they allow estimating visually whether some part of the projection is or is not a reliable image of the original manifolds. It also helps to figure out what the original topology of the data is, telling where the high-dimensional manifolds have been torn or glued during the projection. We experiment these techniques with the principal component analysis and the curvilinear component analysis applied to artificial and real databases.

CheckViz: Sanity Check and Topological Clues for Linear and Non-Linear Mappings

Multidimensional scaling is a must‐have tool for visual data miners, projecting multidimensional data onto a two‐dimensional plane. However, what we see is not necessarily what we think about. In many cases, end‐users do not take care of scaling the projection space with respect to the multidimensional space. Anyway, when using non‐linear mappings, scaling is not even possible. Yet, without scaling geometrical structures which might appear do not make more sense than considering a random map. Without scaling, we shall not make inference from the display back to the multidimensional space. No clusters, no trends, no outliers, there is nothing to infer without first quantifying the mapping quality. Several methods to qualify mappings have been devised. Here, we propose CheckViz, a new method belonging to the framework of Verity Visualization. We define a two‐dimensional perceptually uniform colour coding which allows visualizing tears and false neighbourhoods, the two elementary and complementary types of geometrical mapping distortions, straight onto the map at the location where they occur. As examples shall demonstrate, this visualization method is essential to help users make sense out of the mappings and to prevent them from over interpretations. It could be applied to check other mappings as well.

High-dimensional labeled data analysis with topology representing graphs

We propose the use of topology representing graphs for the exploratory analysis of high-dimensional labeled data. The Delaunay graph contains all the topological information needed to analyze the topology of the classes (e.g. the number of separate clusters of a given class, the way these clusters are in contact with each other or the shape of these clusters). The Delaunay graph also allows to sample the decision boundary of the Nearest Neighbor rule, to define a topological criterion of non-linear separability of the classes and to find data which are near the decision boundary so that their label must be considered carefully. This graph then provides a way to analyze the complexity of a classification problem, and tools for decision support. When the Delaunay graph is not tractable in too high-dimensional spaces, we propose to use the Gabriel graph instead and discuss the limits of this approach. This analysis technique is complementary with projection techniques, as it allows to handle the data as they are in the data space, avoiding projection distortions. We apply it to analyze the well-known Iris database and a seismic events database.

Data-driven Evaluation of Visual Quality Measures

Visual quality measures seek to algorithmically imitate human judgments of patterns such as class separability, correlation, or outliers. In this paper, we propose a novel data‐driven framework for evaluating such measures. The basic idea is to take a large set of visually encoded data, such as scatterplots, with reliable human “ground truth” judgements, and to use this human‐labeled data to learn how well a measure would predict human judgements on previously unseen data. Measures can then be evaluated based on predictive performance—an approach that is crucial for generalizing across datasets but has gained little attention so far. To illustrate our framework, we use it to evaluate 15 state‐of‐the‐art class separation measures, using human ground truth data from 828 class separation judgments on color‐coded 2D scatterplots.

Learning topology of a labeled data set with the supervised generative Gaussian graph

Extracting the topology of a set of a labeled data is expected to provide important information in order to analyze the data or to design a better decision system. In this work, we propose to extend the generative Gaussian graph to supervised learning in order to extract the topology of labeled data sets. The graph obtained learns the intra-class and inter-class connectedness and also the manifold-overlapping of the different classes. We propose a way to vizualize these topological features. We apply it to analyze the well-known Iris database and the three-phase pipe flow database.

ProxiLens: Interactive Exploration of High-Dimensional Data using Projections

As dimensionality increases, analysts are faced with difficult problems to make sense of their data. In exploratory data analysis, multidimensional scaling projections can help analyst to discover patterns by identifying outliers and enabling visual clustering. However to exploit these projections, artifacts and interpretation issues must be overcome. We present ProxiLens, a semantic lens which helps exploring data interactively. The analyst becomes aware of the artifacts navigating in a continuous way through the 2D projection in order to cluster and analyze data. We demonstrate the applicability of our technique for visual clustering on synthetic and real data sets.

SepMe: 2002 New visual separation measures

Our goal is to accurately model human class separation judgements in color-coded scatterplots. Towards this goal, we propose a set of 2002 visual separation measures, by systematically combining 17 neighborhood graphs and 14 class purity functions, with different parameterizations. Using a Machine Learning framework, we evaluate these measures based on how well they predict human separation judgements. We found that more than 58% of the 2002 new measures outperform the best state-of-the-art Distance Consistency (DSC) measure. Among the 2002, the best measure is the average proportion of same-class neighbors among the 0.35-Observable Neighbors of each point of the target class (short GONG 0.35 DIR CPT), with a prediction accuracy of 92.9%, which is 11.7% better than DSC. We also discuss alternative, well-performing measures and give guidelines when to use which.

Sanity check for class-coloring-based evaluation of dimension reduction techniques

Dimension Reduction techniques used to visualize multidimensional data provide a scatterplot spatialization of data similarities. A widespread way to evaluate the quality of such DR techniques is to use labeled data as a ground truth and to call the reader as a witness to qualify the visualization by looking at class-cluster correlations within the scatterplot. We expose the pitfalls of this evaluation process and we propose a principled solution to guide researchers to improve the way they use this visual evaluation of DR techniques.

Implementing 360° Quantified Self for childhood obesity: feasibility study and experiences from a weight loss camp in Qatar

BackgroundThe explosion of consumer electronics and social media are facilitating the rise of the Quantified Self (QS) movement where millions of users are tracking various aspects of their daily life using social media, mobile technology, and wearable devices. Data from mobile phones, wearables and social media can facilitate a better understanding of the health behaviors of individuals. At the same time, there is an unprecedented increase in childhood obesity rates worldwide. This is a cause for grave concern due to its potential long-term health consequences (e.g., diabetes or cardiovascular diseases). Childhood obesity is highly prevalent in Qatar and the Gulf Region. In this study we examine the feasibility of capturing quantified-self data from social media, wearables and mobiles within a weight lost camp for overweight children in Qatar.MethodsOver 50 children (9–12 years old) and parents used a wide range of technologies, including wearable sensors (actigraphy), mobile and social media (WhatsApp and Instagram) to collect data related to physical activity and food, that was then integrated with physiological data to gain insights about their health habits.In this paper, we report about the acquired data and visualization techniques following the 360° Quantified Self (360QS) methodology (Haddadi et al., ICHI 587–92, 2015).Results360QS allows for capturing insights on the behavioral patterns of children and serves as a mechanism to reinforce education of their mothers via social media. We also identified human factors, such as gender and cultural acceptability aspects that can affect the implementation of this technology beyond a feasibility study. Furthermore, technical challenges regarding the visualization and integration of heterogeneous and sparse data sets are described in the paper.ConclusionsWe proved the feasibility of using 360QS in childhood obesity through this pilot study. However, in order to fully implement the 360QS technology careful planning and integration in the health professionals’ workflow is needed.Trial RegistrationThe trial where this study took place is registered at ClinicalTrials.gov on 14 November 2016 (NCT02972164).

How to help seismic analysts to verify the French seismic bulletin

In this paper, classifiers based on Multi-Layer Perceptrons and Support Vector Machines are used in order to classify seismic events that occurred in metropolitan France. The results are exploited in the software RAMSES to help the seismic analysts to conduct efficiently the revision of the weekly French seismic bulletin. With 96.5% of good classification, and less than 7% of the events emphasized for verification, RAMSES strikingly improves the speed of the revision.