Jiangsu Provincial Key Laboratory of E-Business

Enterprise Academician Workstation at FOCUS

Keynote Speakers

Prof. Leslie Valiant
Winner of Turing Award,Harvard University, USA
Title:Extending Machine Learning to Encompass Reasoning
One of the most important challenges for computer science is that of understanding how systems that acquire and manipulate common sense knowledge can be created.

In this talk we shall first formulate this challenge as one of finding a common semantics in which inductive learning and logical reasoning can both be expressed. Learning provides the robustness needed to cope with the uncertainties of the world, providing a mechanism by which the system can go back to the world for further data. Reasoning is needed to provide a principled basis for reaching conclusions in novel situations.

We go on to give a proposed solution to this problem that we call robust logic. This offers a framework, using the semantics of learning, for learning rules that are suitable for later chaining together for the purpose of reasoning. In this system both learning and reasoning can be performed in polynomial time.

Finally we review the results of some experiments, obtained jointly with Loizos Michael, which tested this theory on a dataset of a half a million natural language sentences. The experiments showed that a certain task of predicting a deleted word from a sentence could be performed to a higher accuracy by this method than by a baseline learning method that did not chain together learned rules. As a machine learning exercise these experiments highlight the challenge of learning from noisy data in the exacting setting in which the goal is to learn rules that are reliable enough for chaining together.

Prof.Bo Zhang
Academician of CAS,Tsinghua University, China
Title:Machine Learning and Data Mining
Data mining is to find the regularity behind a huge amount of raw data. Machine learning approaches can be used to deal with the problem generally so there is a close relationship between machine learning and data mining. In data mining or machine learning, one approach is called the cognition-based, i.e., to find a set of rules behind the data by prior knowledge. The other is probability-based approach. Its goal is to find the statistical regularity among the data. In web era, we are confronted with a huge amount of raw data and a tremendous change of man-machine interaction modes. We have to deal with the semantic relations among data rather than their formal relations alone. It is difficult to deal with the semantic problem by the two approaches mentioned above. So it’s needed a new information processing strategy that correlated with the content of information by learning some mechanisms from human beings. Therefore, we need (1) a set of robust detectors for detecting semantically meaningful features among the data, such as boundaries, shapes, etc. in images, words, sentences, etc. in text, and (2) a set of methods that can effectively analyze and exploit the data structures that encode the content of information. During the past 40 years the probability theory has made a great progress. It has provided a set of mathematical tools for representing and analyzing information (or data) structures. In the talk we will discuss what difficulty we face, what we can do, and how we should do in the content-based machine learning (or data mining).

Prof. David Bell
Queen's University Belfast, UK
Title:Towards the Measurement of Knowledge Plasticity
Cognitive Plasticity -the capacity to accumulate,adjust and improvecertain ‘mental’capabilities– is important for solving problems andmanaging information in changing circumstances.It is arequiredcharacteristic of flexible learners and reasoners of all sorts, and in particular it is needed in the journey towards ever less ‘playback’ computing systems.

This talk is aboutwork on providing this capacity in artefacts, and in particular on the search for a means of measuringone aspect of this capacity in the context of volatile and expanding knowledge. Our goal is not to understand human or other organic cognition, but to support the engineering of automated agents. However, by common consent, cognitive processes that are manifested in humans and other speciescan reveal fundamental capabilities that are needed for developing functionality which is ‘pulled’ by practical application needs in computing systems, such as those for knowledge discovery in databases, surveillance, animal behaviour modelling, robots, software agents and other infrastructural support in computing.

A focus of our work is on knowledge accumulation which has a particular pattern. An outline will be presented of an approach to a sort of ‘knowledge dynamics’ for use in this. The emphasis is on a way of measuring changes and rates of change in the knowledge of an agent which adapts to or explores its environment. This is justa component of that wider scheme, but such a method could have applications in its own right.Insights are sought onhowlearnerscan accumulate and deal withconflicting knowledge fromnew things which happen, and other changes which arise, unannounced, in their ‘experience’. The term ‘knowledge plasticity’ is meant to capture this focus.

Prof. Qiang Yang
Huawei Noah's Ark Research Lab, Hong Kong
Title: Transfer learning and Applications
In machine learning and data mining, we often encounter situations where we have an insufficient amount of high-quality data in a target domain,but we may have plenty of auxiliary data in related domains. Transfer learning aims to exploit these additional data to improve the learning performance in the target domain. In this talk, I will give an overview on some recent advances in transfer learning for challenging data mining problems. I will present some theoretical challenges to transfer learning, survey the solutions to them, and discuss several innovative applications of transfer learning, including learning in heterogeneous cross-media domains and in online recommendation, social media and social network mining.

Prof. Ester Martin
Simon Fraser University, Canada
Title: Data Mining Methods For Exploiting the Full Potential of Social Media
Social media are media for social interaction, using highly accessible and scalable communication techniques to create and exchange user-generated content. While conventional media such as newspapers and TV are restricted to professional authors and are expensive to produce, social media are associated with low costs and allow large numbers of amateurs to publish their content. Another distinctive feature of social media, enabled by Web 2.0 technology, is the support of various forms of interaction among content producers and consumers, e.g. by sharing, rating, and commenting on user-generated content. Social media sites provide a rapidly increasing wealth of potentially valuable content, but their huge size makes it very hard for users to find the most relevant content, and their lower quality, compared to conventional media, complicates the automatic processing. Therefore, researchers are exploring novel data mining methods to exploit the full potential of social media.

In this talk, we will provide a brief overview of the main tasks of data mining in social media. We will then take a closer look at a key ingredient of social media, social networks. In the social sciences, the effects of social influence, homophily or selection, and transitivity have been identified as drivers of the dynamics of social networks. This raises the following questions, which we will address: How can these effects been modeled computationally? How can we learn the strength of social influence? How can these effects be exploited for improved recommendation of items and links? In the second main part of the talk, we will explore another key ingredient of social media, namely user-provided content, considering online product reviews as a representative example. How can we classify expressed sentiments, which are so typical for social media? How can large collections of product reviews be summarized in a meaningful way? How can we assist users by recommending the most helpful reviews? The talk will discuss our own work as well as important work from the literature, with a focus on probabilistic graphical methods. As the conclusion, we will outline interesting directions for future research in the fascinating field of data mining in social media.

Important Due Dates

Workshop Proposals Due:
May 10, 2012

Abstract Submission Deadline:
July 10, 2012,July 31, 2012

Full Paper Submission Deadline:
July 17, 2012,July 31, 2012

Notification of Acceptance:
September 16, 2012,September 30, 2012

Camera Ready Submission Due:
September 30, 2012,October 14, 2012