Special Issue on Multi-instance Learning in Pattern Recognition and Vision
Aim and Scope
Multi-instance learning (MIL) has served as an important tool for a wide range of applications in patter recognition and computer vision, for instance, drug activity prediction, text classification, image classification, object detection, and visual tracking. In recent years, understanding big visual data in mobile internet is a new trend. With the development of deep neural networks, the performances of many large-scale visual recognition tasks have been significantly improved. However, supervised deep learning methods, e.g., deep convolutional neural net- works (DCNN), rely heavily on the huge number of human-annotated data that are non-trivial to get. Finely labeled images/videos, which have pixel labels and bounding-box labels, are very limited and expensive. However, there are tons of weakly labeled visual data that have image-level labels. For example, we can easily obtain a rough label of an image from its text caption on Flickr. Similarly, the amount of weakly labeled data is much larger than finely labeled data when analysing text, audio and other media data. How to use the weakly labeled media data for media understanding is an important research problem.
Multi-instance learning (MIL), proposed by Dietterich et al. for the purpose of drug activity prediction, is a popular tool for exploring semantic information in weakly labeled data. In MIL, instead of being given the labels of each individual instance, the learner receives a set of labeled bags, each containing plenty of instances. MIL algorithms can not only learning bag classifier, but also inferring the semantic label of instances to find object of interested, for example, locating objects in image, tracking objects in video, or finding key words in a paragraph.
Some important progress has been made in this aspect, which shows MIL is becoming more and more powerful in real applications. For example, Li Fei-Fei et al. use MIL for weakly supervised object location on ImageNet, Cordelia Schmid et al. follow a MIL approach that iteratively trains the detector and infers the object locations, Pedro Felzenszwalb et al. use a MIL method named latent SVM for learning part-based object detector (DPM) which is robust to object deformation, and the researchers from Microsoft Research and Facebook Research use MIL with DCNN to learn visual concepts and image captions from big weakly labeled image data.
However, to our knowledge, no special issue in an international journal de- voting to this important topic. Now it is the right time to organize a special issue in Pattern Recognition to promote the developments of MIL theories and applications in pattern recognition and computer vision, which attracts much attention from both academia and industry. We hope this topic will aggregate high quality works on the new advances in MIL. We will solicit original contributions of researchers and practitioners from the academia as well as industry, which address a wide range of theoretical and applied issues.
The topics of interest include, but are not limited to:
Weakly-supervised object localization/common object discovery
MIL for object detection/tracking,
MIL for image classification/annotation
New MIL algorithms
MIL with deep learning
Key instance detection in MIL
Multi-label multi-instance learning
Text/document classification with MIL
MIL for natural language processing
First submission date: Mar. 1, 2016
Submission deadline: May 1, 2016
First review notification: Jun. 20, 2016
Revised submission due: Aug. 1, 2016
Notification of second-round review: Sep. 10, 2016
Final round of revisions: Oct. 10, 2016
Final paper notification: Dec. 1, 2016
Camera-ready due: Dec. 15, 2016
Instructions for Submission
The submission website for this journal is located at: http://ees.elsevier.com/pr/default.asp
To ensure that all manuscripts are correctly identified for inclusion into the special issue you are editing, it is important that authors select “SI:MIL-PRV” when they reach the “Article Type Name” step in the submission process.
Dr. Jianxin Wu (firstname.lastname@example.org), Nanjing University, Nanjing, China
Dr. Xiang Bai (email@example.com), Huazhong University of Science and Technology, Wuhan, China
Dr. Marco Loog (M.Loog@tudelft.nl), Delft University of Technology, The Netherlands
Dr. Fabio Roli (firstname.lastname@example.org), University of Cagliari, Italy
Dr. Zhi-Hua Zhou (email@example.com), Nanjing University, Nanjing, China