Tutorial Chairs

Anastasia Ailamaki

Ecole Polytechnique Federale de Lausanne, Switzerland

• 
• 

Ioana Manolescu

INRIA, France

• 
• 

• Continuous Measurement of Quality of Data Streams

  Tamraparni Dasu, Rong Duan and Divesh Srivastava (AT&T Labs, USA)
  Abstract +-

  Monitoring complex, dynamic phenomena, e.g., movement of financial markets and traffic in communication infrastructures, produces highly detailed, evolving data streams. Glitches in the monitoring and data delivery infrastructure can lead to a variety of quality problems in the streams, including lost, late or out-of-order data delivery, and erroneous observations. Since data streams are often continuously analyzed to enable real-time, data-driven decision making, it is imperative to continuously measure and alert on the quality of the data streams as well. This tutorial presents (i) an end-to-end framework for data quality measurement in a dynamic environment of evolving data streams, (ii) a variety of recent research from the data management and data mining communities within this framework, and identifies a range of open problems in this area.

  
  

• Model Selection and Error Estimation without the Agonizing Pain

  Luca Oneto and Davide Anguita (University of Genoa, Italy)
  Abstract +-

  How can we select the best performing predictive model? How can we rigorously estimate its generalization error? Statistical Learning Theory (SLT) answers these questions by deriving non-asymptotic bounds on the generalization error of a model or, in other words, by upper bounding the true error of the learned model based just on quantities computed on the available data. However, for a long time, SLT has been considered only an abstract theoretical framework, useful for inspiring new learning approaches, but with limited applicability to practical problems. The purpose of this tutorial is to give an intelligible overview of the problems of Model Selection and Error Estimation, by focusing on the ideas behind the different SLT-based approaches and simplifying most of the technical aspects with the purpose of making them more accessible and usable in practice. We will start by presenting the seminal works of the 80’s until the most recent results, then discuss open problems and finally outline future directions of this field of research.

  
  

• Similarity Search on Time Series Data: Past, Present and Future

  Abdullah Mueen (University of New Mexico, USA)
  Abstract +-

  Similarity search on time series data has been studied for decades, and we continue to see new solutions being developed to address challenges rising from novel applications, natural constraints, and processing platforms. In this tutorial, we briefly describe the last three decades of research on similarity search in time series data, discuss recent developments and project future challenges towards similarity based temporal mining systems. The tutorial will cover a large set of similarity measures, domain dependent invariances, bounding techniques for indexing and scalability, and representational techniques for efficiency. We will discuss relative strengths and weaknesses of various methods and make recommendations for researchers and practitioners. We will present high-level mining problems that directly benefit from similarity search and are still open. The tutorial will discuss a rich set of application areas where similarity search has led to successful knowledge discovery, such as physiological monitoring, data center monitoring, activity recognition and animal tracking.