HighStream’2017
High-Performance Data Stream Mining Workshop
co-located with IEEE International Conference on Data Mining ICDM 2017

Learning from data streams have emerged as one of the most vital topic 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. Volume and velocity are difficult tasks to handle on their own, yet they need to be considered from a perspective of non-stationary problems affected with a phenomenon known as concept drift. This problem has been thoroughly studied in the last decade with a specific focus on classification tasks. However, the research community has started to address this problem within other contexts such as data preprocessing, regression, multi-label classification, association rule mining, imbalanced learning, graph and xml mining, social and mobile networks, as well as novelty detection. It is now recognized that imbalanced domains are a broader and important problem posing relevant challenges for both supervised and unsupervised learning tasks, with handling various embedded difficulties in an increasing number of real world applications.

Tackling the issues raised by data stream mining is of high importance to people from both academia and industry. For researchers, these challenges offer an exciting option to develop adapting, evolving and efficient learning methods that will be able to handle such difficult cases. For industry, many of real problems to be faced actually arrive in form of streams, thus such methods are vital for tackling these tasks. They require methods that enable a more preemptive, real-time action in an increasingly fast-paced world and are able to constantly update and evolve knowledge and models in accordance with the current state of data. Additionally, with the ever-increasing scale and complexity of these problems, we need high-performance computing environments (clusters, cloud computing, GPUs) and fast, incremental, ideally single-pass algorithms to offer highest possible predictive power at lowest time and computational cost.

The research topics of interest to HighStream'2017 workshop include (but are not limited to) the following:

+ Foundations of learning from data streams
Probabilistic and statistical models
Understanding the nature of learning difficulties embedded in streaming and non-stationary data
Identifying and handling concept drift
High-performance computing environments for big data streams
Deep learning with streaming data
New approaches for data pre-processing (e.g. discretization strategies)
Post-processing approaches
Sampling approaches
Feature selection and feature transformation
Evaluation in streaming domains
Online model selection
Learning for heterogeneous and multiple data streams
Context-awareness for data stream mining
Resource-aware learning from data streams
Knowledge discovery and data mining in data streams
Classification
Regression
Clustering
Novelty detection and evolving class structures
Learning from imbalanced data streams
Active learning and label latency
Multi-label, multi-instance, sequence and association rules mining
Graph stream mining
On-line ensemble models
Smart data mining with compact models
Applications in solving real-life problems
Social network applications
Medical data streams
Ubiquitous and mobile stream mining
Engineering and industrial applications
Fraud and intrusion detection
Environmental applications