Data Distillation for Traffic Sign Detection

This work is devoted to the traffic sign detection on images using deep learning methods. We focus on the problem of detector transfer to new datasets with different road signs. We present an algorithm for distilling a set of unlabelled data to select the most informative images to be labeled. This method allows to significantly reduce the amount of data labeling with a small decline of detector performance.

DDFlow: Learning Optical Flow with Unlabeled Data Distillation

We present DDFlow, a data distillation approach to learning optical flow estimation from unlabeled data. The approach distills reliable predictions from a teacher network, and uses these predictions as annotations to guide a student network to learn optical flow. Unlike existing work relying on handcrafted energy terms to handle occlusion, our approach is data-driven, and learns optical flow for occluded pixels. This enables us to train our model with a much simpler loss function, and achieve a much higher accuracy. We conduct a rigorous evaluation on the challenging Flying Chairs, MPI Sintel, KITTI 2012 and 2015 benchmarks, and show that our approach significantly outperforms all existing unsupervised learning methods, while running at real time.

Translating Research into Agile Development (TRIAD): Development of Electronic Health Record Tools for Primary Care Settings

Objectives This article describes a method for developing electronic health record (EHR) tools for use in primary care settings. Methods The “Translating Research into Agile Development” (TRIAD) method relies on the close collaboration of researchers, end users, and development teams. This five-step method for designing a tailored EHR tool includes (1) assessment, observation, and documentation; (2) structured engagement for collaboration and iterative data collection; (3) data distillation; (4) developmental feedback from clinical team members on high-priority EHR needs and input on design prototypes and EHR functionality; and (5) agile scrum sprint cycles for prototype development. Results The TRIAD method was used to modify an existing EHR for behavioral health clinicians (BHCs) embedded with primary care teams, called the BH e-Suite. The structured engagement processes stimulated discussions on how best to automate BHC screening tools and provide goal tracking functionality over time. Data distillation procedures rendered technical documents, with information on workflow steps, tasks, and associated challenges. In the developmental feedback phase, BHCs gave input on screening assessments, scoring needs, and other functionality to inform prototype feature development. Six 2-week sprint cycles were conducted to address three domains of prototype development: assessment and documentation needs, information retrieval, and monitoring and tracking. The BH e-Suite tool resulted with eight new EHR features to accommodate BHCs' needs. Conclusion The TRIAD method can be used to develop EHR functionality to address the evolving needs of health professionals in primary care and other settings. The BH e-Suite was developed through TRIAD and was found to be acceptable, easy to use, and improved care delivery during pilot testing. The BH e-Suite was later adopted by OCHIN Inc., which provided the tool to its 640 community health centers. This suggests that the TRIAD method is a promising research and development approach.