Latest news:


Welcome to the 3rd IEEE BDTL!
Knowledge Transfer and Multi-view Learning (KTMvL)

Although widely applied on considerable scientific research, conventional statistical machine learning revolves on a simplified assumption that the training data, from which the algorithms learn, are drawn i.i.d. from the same distribution as the test data, to which the learned models are applied. This assumption, being broken down by numerous real-world applications nowadays, especially with the emergence of large-scale data from the private internal data, or the public Internet, has fundamentally restricted the development of practical learning al-gorithms. More examples include but are not limited to: (1) speech recognition when speakers have strong dialects from different countries, regions, genders, or aging groups; (2) surveillance system where captured suspects' faces are side view, under unknown lighting conditions, in low-resolutions, different from the conditions of the registered faces in the system.

On the other hand, multi-view data generated from various view-points or multiple sensors are commonly seen in real-world applications. For example, the popular commercial depth sensor Kinect uses both visible light and near infrared sensors for depth estimation; autopilot uses both visual and radar sensors to produce real-time 3D information on the road; face analysis algorithms prefer face images from different views for high fidelity reconstruction and recognition. However, such data with large view divergence would result in an enormous challenge: data lying in different views show a large divergence preventing them from a fair comparison. In general, different views can be treated as different domains drawn from different distributions. Therefore, there is an urgent need to mitigate the view divergence by either fusing the knowledge across multiple views or adapting knowledge from some views to others.

The key problems discussed above come down to two popular research topics in modern data mining: transfer learning and multi-view learning. While the first problem emphasizes ex-ploring sample-wise correspondence across different views, the second problem focuses more on the generic knowledge transfer or adaptation, e.g., intelligent recognition. Essentially, they both attempt to address the issues of knowledge transfer or fusion between different domains, which makes significant sense given large amount of available auxiliary data from other datasets, sensors, or modalities.