VUMLA 2014 - International Workshop on Visualizing Uncertainty in Machine Learning Applications

VUMLA 2014 is the 1st International Workshop dealing with Uncertainty Visualization in Machine Learning Applications. It is held in conjunction with the International Working Conference on Advanced Visual Interfaces (AVI 2014), May 27-30 2014, Como, Italy.
Machine learning technologies have opened a wide range of new practices in a variety of domains, some of them being key assets for society. Transferring these technologies to an application domain can lead to trust issues, since machine learning results contain uncertainty. There are multiple factors impacting the reliability of the information derived from machine learning solutions, and each domain of application has specific requirements regarding uncertainty. The VUMLA Workshop addresses the challenges of using visualization to convey uncertainty for specific application needs.
Visualization can enable end-users to investigate not only the results of machine learning solutions, but also the associated uncertainty. It can support the uptake of machine learning outcomes, and avoid distrust or rejection of valid solutions, misinterpretation of results, and moreover, safety issues (e.g., in medicine or surveillance). Yet dealing with uncertainty is challenging for both users and computer scientists. Users need to evaluate the reliability of available solutions, and the information they supply, while having no expertise in the underlying techniques. The methods for evaluating machine learning techniques are primarily designed by and for machine learning experts. They are typically not easy to understand by end-users. Moreover, some uncertainty factors may not be relevant (e.g., pixel-level evaluation in computer vision), and additional uncertainty factors may not be considered (e.g., specific quality and proportions of ground-truth items, uncertainty propagation along information processing). There is a need for providing uncertainty visualizations that expose the applicable technical limitations, evaluating either the general reliability of machine learning algorithms (e.g., in test conditions), or the resulting datasets (e.g., in end-usage conditions).
The VUMLA workshop gathers a community of researchers and practitioners interested in uncertainty issues, particularly in the context of machine learning applications and uncertainty-aware visualization design. The presented papers and the discussion session deliver insights into the human and technical factors involved in the use of uncertain data. It informs and inspires future designs of uncertainty-aware visualizations for machine learning applications. Authors and participants from a variety of backgrounds are encouraged to join the workshop, e.g., users, designers, machine learning experts, or domain experts.