The International Workshop on Active Learning for Big Data (ALBD-2017)

Active Learning addresses the interaction between Machine Learning/Data Mining algorithms and human feedbacks, reduces the human efforts for manual labelling, avoids data redundancy, and improves the computation speed of Machine Learning tools. It has been studied for many years under the traditional single-instance and single-label settings, where each data point is dependent of the others and is belonging to a specific class. On one hand, these learning methods are not applicable to complex scenarios, such as multi-instance and multi-label settings. On the other hand, with the rapid expansion of existing data, there are still gaps between theoretical research and practical applications. 

When designing Active Learning methods for complex scenarios, new issues are raised, including the design of multi-instance or multi-label learners, feature selection methods, sample selection indices, stopping criteria, and performance evaluation metrics, etc. In order to adapt Active Learning to big data problems, methods must be able to handle data with high volumes and high-dimension, with the ability of mining useful information from increasingly large data streams. This workshop aims to provide a forum for researchers to discuss the above-mentioned problems for Active Learning, identify challenges for Active Learning in complex scenarios, provide solutions to Active Learning regarding big data, as well as discover the potentials of Active Learning to new real-world applications. We encourage any related topic for theoretical analysis, methodology design, and real-world applications regarding Active Learning.