KDD17 Tutorial: Non-IID Learning in Big Data

Non-IID learning/Non-IIDness learning in big data refers to the methodologies, algorithms and practical tools for representing, modeling, analyzing and understanding non-IID (not independent and identically distributed) data. The key is to represent, learn and synthesize the intrinsic heterogeneity and coupling relationships (non-IIDness) by developing state-of-the-art data representation methods and learning models. This tutorial introduces the methodologies on the representation of non-IIDness in big data and presents the state-of-the-art techniques and algorithms on incorporating non-IIDness into a variety of learning tasks, including classification, clustering, ensemble clustering, outlier detection, feature selection, recommender systems, and text mining.



Learning from big data is increasingly becoming a major challenge and opportunity for big business and innovative learning theories and tools. Some of the most critical challenges of learning from big data are the uncovering of the explicit and implicit coupling relationships embedded in mixed heterogeneous data from single/multiple sources. The coupling and heterogeneity of the non-IIDness aspects form the essence of big data and most real-world applications, namely the data is non-IID.


Most of classic theoretical systems and tools in statistics, data mining, database, knowledge management and machine learning assume the independence and identical distribution of underlying objects, features and values. Such theories and tools may lead to misleading or incorrect understanding of real-life data complexities. Non-IID learning in big data is a foundational theoretical problem in AI and data science, which considers the couplings and heterogeneity between entities, properties, interactions and contexts. In this tutorial, we present a comprehensive overview of the non-IID learning. We begin with the limitation of IID learning in handling big data and introduce abstract learning model and representation for non-IID learning, and then present frameworks and algorithms for non-IID metric learning, classification, clustering, ensemble clustering, outlier detection, feature selection, recommender systems, and text mining, and finally discuss open challenges and prospects.



The tutorial includes the following contents:

  • The main challenges in Big Data: a brief introduction of Big Data and the essential theoretical challenges in learning Big Data;
  • What is IID Learning: summarize the classic IID learning theories by several examples such as clustering and recommender system theories;
  • Limitations of IID Learning: summarize the main limitations of the classic IID learning methods, and introduce why they do not work well in handling Big Data-oriented problems;
  • Related work of non-IID Learning: introduce the related work of Non-IID learning to address the gaps and issues in classic IID and non-IID data learning theories and models;
  • Non-IIDness characteristics and nature: discuss the characteristics of non-IID data, especially the coupling relationships and heterogeneity in big data;
  • Abstract non-IID learning model and representation: discuss an abstract representation model of non-IIDness by involving the heterogeneity and couplings between entities, properties, contexts and interactions etc.;
  • non-IID clustering: new similarity metrics, algorithms and case studies will be presented to introduce non-IIDness into clustering;
  • non-IID ensemble clustering: new similarity metrics, algorithms and case studies will be presented to cater for non-IIDness in ensemble clustering;
  • non-IID outlier detection: new methods will be introduced to incorporate non-IIDness into outlier detection and outlying feature selection.
  • non-IID classification: similarity metrics and algorithms will be introduced to incorporate non-IIDness into classification;
  • non-IID recommender systems: non-IID recommender systems including the framework, algorithms and experiments will be introduced;
  • Term coupling-based document analysis: similarity metrics, algorithms and case study of incorporating non-IID learning into document analysis will be discussed;
  • Hierarchical structure and couplings in non-IID learning: We summarize the case studies and conclude the hierarchical structure and couplings in non-IID learning as a foundation for non-IID learning;
  • Challenges and prospects of non-IID information processing and management: open issues and opportunities will be discussed for incorporating non-IIDness into information processing and management, including database management, knowledge management, information retrieval, recommender system, data mining and machine learning to develop corresponding methodologies, systems and algorithms for real-world big data applications.

Tutorial Slides  [pdf]

Intended audience:

Non-IID data learning is a fundamental issue in data-driven knowledge discovery and data analytics. Any audience who may be interested in data-driven discovery and data analytics theories and systems, for instance knowledge representation and management, machine learning, data mining, image processing, computer vision, recommender system, online network analysis, social media analysis, and text mining, would find it very inspiring and valuable in attending this tutorial. In this tutorial, the attendant will not only learn the basic knowledge of non-IID learning and Big Data, but also can benefit from the detailed theories and algorithms for solving the IID learning challenges in the above areas and domains.


While no specific knowledge is required from the audience, people who are familiar with the above mentioned areas will find it more beneficial in understanding the algorithms and case studies to be introduced in this tutorial.




Some relevant activities on non-IID learning