CVPR 2017 Workshop on Continuous and Open-Set Learning


The call for papers is currently open for the Continuous and Open-Set Learning workshop at CVPR 2017.  https://erodner.github.io/continuouslearningcvpr2017/

Recent breakthroughs in our community have relied on the availability of large representative datasets for training. However, the implicit assumption imposed in the majority of our today’s techniques is a static closed world, i.e., non-varying distributions for a fixed set of categories and tasks. Intuitively, these assumptions rarely hold in many application areas such as concept detection in biomedical image analysis, explorative data-driven science, scene parsing for autonomous driving, or household robotics. Instead, the set of semantic concepts and relevant tasks is dynamically changing – even on a daily basis. The assumption of a closed and static world is therefore one of the major obstacles when building intelligent systems that learn continuously, adaptively, and actively.

In general, this workshop tries to bridge one of the gaps between computer vision research and AI goals by focusing on different aspects of continuous and open-set learning. In consequence, the following topics will be central to the workshop:

  • Dealing with partially unknown, open, or dynamically increasing label spaces (probabilistic models, possibility for rejection, novelty detection, etc.)
  • Continuous, online, and incremental learning (at level of instances, classes, common-sense knowledge, and representations)
  • Active acquisition and annotation of new data with humans in the loop (curriculum learning, active learning, etc.)
  • Transfer learning and domain adaptation in continuous and open-set learning scenarios
  • Active data discovery in explorative data science and large-scale microscopy data
  • Benchmarking success in continuous and open-set learning scenarios