OPEN VIS CONF 2018 reviewed: reshaping the way we look at data
From machine learning to data design neutrality: between May 14th and May 17th, this emlyon-supported international conference brought the latest in data visualization to Paris.
If you’re into maps or charts, drawing OPEN VIS CONF 2018’s worldwide audience would be fun: the gathering held in Paris between May 14th and May 17th attracted 350 attendees from 26 countries, including Gabon, Chile and Australia, 180 companies and 30 universities. And data visualization, or dataviz, is exactly what this international conference, first staged in Boston in 2013, is all about. Visualizing data through charts, maps, dashboards or pretty much anything to showcase sorted information and “educate, raise attention, spark curiosity or support analysis”, to quote keynote speaker Caroline Goulard, CEO of Dataveyes. From data scientists to web developers, analytics consultants to graphic designers, meteorologists to bar chart geeks, the conference certainly sparked curiosity. As datasets continue growing steadily over the years, choosing the right ways to present or monitor information is obviously crucial.
While dataviz conferences are increasingly common in North America and Asia, they are still quite rare in France and Europe. OPEN VIS CONF 2018 was made possible by the financial and organizational support of emlyon, a business school with strong connections to data science. After setting up its first data R&D institute in 2017, emlyon plans to bring together business analytics and data science this year by launching its MSc in Data Science and Digital Marketing in fall 2018. Bridging business analytics and data science may be an ambitious challenge, but OPEN VIS CONF 2018 took it on, offering participants comprehensive insights, cool animations and deep questions about the latest trends in data design. Boasting a diverse line-up of speakers from innovative companies (Google, Tulp, Qlik), cutting-edge media (The New York Times) and prestigious schools (MIT, University of Utah), the conference set the stage to find inspiration, learn about new tools and explore dataviz ethics. And of course attendees also had plenty of opportunity between talks to network and identify new business opportunities.
Media & public policy: make the invisible visible
Data is not a new field, when you consider that William Playfair and Charles Minard, who are regarded as the first two modern “data visualizers”, were working back in the 19th century. But it has grown hugely over the last 20 years, and production facilities and studios are springing up everywhere. It would be tedious to review every single project displayed at this year’s conference, but the topics covered ranged from visualizing climate change in a dynamic and interactive way as shown by The New York Times’ graphics editor Nadja Popovich to visualizing racism in the US and 3D science stories. The Google Sunroof project calculates how much money you can save by investing in solar panels, while the Dataveyes Compagnon app helps users improve energy consumption through smart electricity meter monitoring.
What do all these projects have in common? They “make the invisible visible”, explains Caroline Goulard, CEO of Dataveyes. That includes things that we either don’t want to see or simply can’t see, such as noise distribution in the Paris subway, or underlying patterns impossible to catch with a simple excel sheet.
Machine learning and data visualization: a profitable combo
A key topic addressed at OPENVIS CONF 2018 was machine learning’s growing importance in (re)defining data visualization. Much mention was made of the t-SNE algorithm, which can be used to visualize multidimensional and complex datasets learned and analyzed by artificial neural networks. In his talk on “Machine Learning for Visualization”, Google’s Ian Johnson pointed out how t-SNE can help navigate through the data and how a “clustering approach” can make unexpected patterns pop up. And applied that beautifully to the Google’s Quick Draw dataset…
Data visualization can do some unexpected things too, such as making machine learning processings easier to understand, as brilliantly shown by Google’s Shan Carter in his “Lessons From a Year Distilling Machine Learning Research” presentation, taken from his distill.pub project. Algorithms and neural network mechanics can be something of a black box, impenetrable to almost everyone, but Shan Carter showed how visual explanations can help demystify these powerful tools for a larger audience.
Unrelated but similarly avant-gardist was Benjamin Bach’s talk on “Designing Data Visualizations for Augmented Reality”. While AR has yet to become common in our lives, Dr Bach, a lecturer at the University of Edinburgh, looks at the possibilities and design issues stemming from embedding data representations in AR programs. Although this research remains at the theoretical level for now, it could ultimately help us in a range of ways, from monitoring public transport to accessing useful, geolocalized information in bookstores.
Dataviz is not neutral: ethics and cognitive bias
This question cropped up regularly during the talks: though data visualization may often seem impartial, it cannot be neutral. The way it is designed clearly influences how it is read or perceived. To steer clear of ethical or cognitive traps, data designer must think clearly and thoroughly before setting up graphs or dashboards. Heather Klause from Dataassist provided a good explanation of how to “Protect your work from four hidden fallacies when working with data”. She described how the way questions are worded, the variables used to present data and the assumptions made can reshape the story told by the data in graphics. Everyone needs to be aware of this and take onboard the emerging set of best practices and principles in the world of data.
Steven Franconeri, from Northwestern University, gave a thought-provoking lecture on visual cognitive bias. Data visualizations are deeply ambiguous figures, and our brain has at least three tools to understand even a simple two-bar graph: object recognition, which happens fast, feature distribution (also fast) and comparison (very slow). It may seem far-fetched, but the slight delay in visual processing can have important consequences for how a chart or graph is understood: it’s crucial to disambiguate data visualization.
Business opportunities: redesigning dashboards !
Everyone knows you need to master your data to enable your business to optimize processes, improve customer relationships and/or generate revenues based on better analytics-sourced decisions. In a data market estimated at €124.5 billion worldwide, though, mining data is one thing, visualizing and monitoring data more efficiently is another.
A number of talks tackled this topic from the business side. Some were essentially inspirational, like the presentation by Maarten Lambrechts, whose Xenographics project collects “graphics that people haven’t seen before” — for better or for worse.
Conversely, the first keynote speaker, Moritz Stefaner, a German data visualization specialist and freelance “Truth and Beauty Operator”, gave a nice example of how dataviz can be used professionally. After a enlightening introduction to his t-SNE based visual map of Paris, which was exhibited at Fondation EDF as part of the 1,2,3 Data exhibition (along with his famous food-based “data cuisine” experiments), he presented his latest work with Deutsche Bahn.
This project involved redesigning the monitoring dashboards used by the German railway system to “maximize yield management, fluidify traffic and create predictive models of passenger loads”. From austere Excel spreadsheets, he managed to craft a ux-friendly dashboard that was data-driven, user-centric and beautiful to use, creating the sort of decompartmentalized, easy-to-use system that all practitioners should be striving to achieve.
From data journalism to dashboard design, from public policy management to meteorology, the practical implementations of dataviz are endless. All you have to do is ask the right questions.
Useful links
