Introduction
The International Workshop on Adaptive and Deep Learning Approaches for Data Streams is to be held in conjunction with 19th IEEE International Conference on Data Mining. Beijing, China, 8-11 November 2019.
Learning from data streams have emerged as one of the most vital topics in contemporary machine learning and data stream mining. They encompass several challenges for modern intelligent systems: potentially unbounded volume of data, instances arriving at high speed in varying intervals, changing and evolving decision space, difficulties with access to ground truth, as well as need for managing heterogeneous forms of information. A major challenge in data stream mining is how to use adaptive learning in order to cope with the encountered change such as concept drift, concept evolution and feature evolution. These changes cause a difference between deployment and target domains, hence the mining performance is negatively impacted. Addressing these challenges requires novel adaptive and transfer learning approaches that are designed specially for data streams.
As an extremely active research area in recent years, deep learning has achieved great success in many fields, such as computer vision, natural language processing and speech recognition, through its outperforming learning capability from massive amounts of data. The new developments in deep learning can be used to advance learning from data streams to detect and cope with encountered challenges. However, applying existing deep learning methods to data streams, especially high-speed non-stationary data streams, is not straightforward. First, a DNN model tends to converge slower than a shallow NN, when processing a small sample size problem. Second, training a DNN relies on an iterative parameter learning scenario where the tuning phase is iterated across a number of epochs.
The aim of this workshop is to bring together researchers from the areas of stream mining, deep learning and adaptive learning in order to encourage
discussions and new collaborations on solving the combined issue of adaptive and deep learning in data streams. In order to advance the state-of-the-art on
the combined issue, it is important to also advance the state-of-the-art in each individual area. Therefore, this workshop encourages submissions not only on the combined topic, but also on each individual area. This workshop will provide a forum for international researchers and practitioners to share and discuss their original work on addressing new challenges and research issues in adaptive and deep learning for stream mining.
Learning from data streams have emerged as one of the most vital topics in contemporary machine learning and data stream mining. They encompass several challenges for modern intelligent systems: potentially unbounded volume of data, instances arriving at high speed in varying intervals, changing and evolving decision space, difficulties with access to ground truth, as well as need for managing heterogeneous forms of information. A major challenge in data stream mining is how to use adaptive learning in order to cope with the encountered change such as concept drift, concept evolution and feature evolution. These changes cause a difference between deployment and target domains, hence the mining performance is negatively impacted. Addressing these challenges requires novel adaptive and transfer learning approaches that are designed specially for data streams.
As an extremely active research area in recent years, deep learning has achieved great success in many fields, such as computer vision, natural language processing and speech recognition, through its outperforming learning capability from massive amounts of data. The new developments in deep learning can be used to advance learning from data streams to detect and cope with encountered challenges. However, applying existing deep learning methods to data streams, especially high-speed non-stationary data streams, is not straightforward. First, a DNN model tends to converge slower than a shallow NN, when processing a small sample size problem. Second, training a DNN relies on an iterative parameter learning scenario where the tuning phase is iterated across a number of epochs.
The aim of this workshop is to bring together researchers from the areas of stream mining, deep learning and adaptive learning in order to encourage
discussions and new collaborations on solving the combined issue of adaptive and deep learning in data streams. In order to advance the state-of-the-art on
the combined issue, it is important to also advance the state-of-the-art in each individual area. Therefore, this workshop encourages submissions not only on the combined topic, but also on each individual area. This workshop will provide a forum for international researchers and practitioners to share and discuss their original work on addressing new challenges and research issues in adaptive and deep learning for stream mining.
Topics of interest
- Data stream classification
- Unsupervised learning from data streams (clustering and outlier detection)
- Semi-supervised and transfer learning approaches for learning from data streams
- Class imbalance problems in data streams
- Adaptive learning approaches, including concept drift, concept evolution and feature evolution algorithms
- Deep learning techniques for non-stationary data streams
- Zero-shot and few-shot learning in deep learning
- Dealing with streaming adversarial attacks in deep learning systems
- Hybrid methods between stream mining and deep learning
- Scalable data stream mining
- Case studies and real-world applications
Important dates
- Paper submission: August 7 2019 -> extended to August 24 2019 (new deadline!)
- Paper notification: September 4 2019
- Camera-ready deadline and copyright forms: September 8 2019
- Date of Workshop: November 8-11, 2019 (The exact day to be confirmed)
Workshop chairs
- Shuo Wang, Birmingham City University, UK
- Zahraa Abdallah, Birmingham City University, UK
- Mohamed Medhat Gaber, Birmingham City University, UK
Paper submission and publication
All papers should conform to ICDM main track formatting requirements at http://icdm2019.bigke.org. Papers should be submitted electronically (in PDF) through the ICDM online submission system (also can be found at http://icdm2019.bigke.org/). Authors should select the workshop "ADLADS" from the drop down list for submission. Authors should only submit original work that has neither appeared elsewhere for publication, nor which is under review for another publication.
At least one author from each accepted paper must register for the workshop. All accepted workshop papers will be published in the IEEE Computer Society Digital Library (CSDL) and IEEE Xplore, and indexed by EI.
At least one author from each accepted paper must register for the workshop. All accepted workshop papers will be published in the IEEE Computer Society Digital Library (CSDL) and IEEE Xplore, and indexed by EI.