Background and Motivation

Traditional supervised learning works under the single-label scenario, i.e. each example is associated with one single label characterizing its property. However, in many real-world applications, objects are usually associated with multiple labels simultaneously. One natural example is text categorization, where each document may belong to several predefined topics, such as Shanghai World Expo, economics and even volunteers.

In multi-label learning, each example in the training set is associated with a set of labels and two main tasks are: a) to predict the label set of unseen examples, and b) to rank all labels according to relevance with unseen examples, through analyzing training examples with known label sets. Following the earlier work on multi-label text categorization since 1999, multi-label learning has gradually attracted more and more attentions from machine learning and other related communities.

Specifically, the ever-increasing interest on learning from multi-label data is witnessed by the remarkable amount of works on: a) novel multi-label learning algorithms; b) applications of multi-label learning techniques such as (semi)automated annotation of large collections of image/video, music, text, web and biology objects, drug discovery, query categorization, medical diagnosis, tag recommendation, and direct marketing; c) learning tasks related to multi-label learning, such as dimensionality reduction for multi-label data, hierarchical multi-label learning, semi-supervised multi-label learning, active multi-label learning, multi-instance multi-label learning; and others.

However, despite the encouraging progress in multi-label learning research, there are still many open issues to be addressed in this emerging learning scenario. Following the success of MLD'09 (First International Workshop on Learning from Multi-Label Data), we aim to organize another edition of this workshop to provide an open and interactive forum for people with diverse backgrounds that are interested in multi-label learning, to share their expertise, exchange the ideas and discuss on related issues.

Topics of Interest

This workshop focuses on advancing the research on multi-label learning. To achieve this goal, we solicit good contributions from the following non-exhaustive list of interested topics:

*

Theoretical analysis of multi-label learning
*

Novel methodologies/algorithms to learn from multi-label data
*

Scalable multi-label learning (presence of large number of labels)
*

Learning and/or exploiting label structure (constraints, hierarchies, ontologies)
*

Learning from data with multiple non-binary target variables (nominal, real-valued, mixed)
*

Evaluation in multi-label learning
*

New applications of multi-label learning
*

Related learning tasks
o

Feature selection from multi-label data
o

Multi-instance multi-label learning
o

Active multi-label learning
o

Semi-supervised multi-label learning

**Note that the best selected papers of MLD'10 may be invited to submit an extended version to a special issue at Machine Learning Journal on the topic of Learning from Multi-Label Data. Participants of this workshop are also encouraged to submit their high-quality works to this special issue.**