Keynote Lectures

Advances in Learning with Bayesian Networks
Philippe Leray, Université de Nantes, France

The New Era of High-Functionality Computing
Henry Lieberman, Independent Researcher, United States

Agents and Semantics for Human Decision-Making
Matthias Klusch, German Research Center for Artificial Intelligence (DFKI) GmbH, Germany

The Brain Agent
Claude Frasson, University of Montreal, Canada

The Intelligence of Agents in Games
Pieter Spronck, Tilburg University, Netherlands

Demystifying Big Data - From Causality to Correlation
Jaap van den Herik, Leiden University, Netherlands

Abstract

Bayesian networks (BNs) are a powerful tool for graphical representation of the underlying knowledge in the data and reasoning with incomplete or imprecise observations.
BNs have been extended (or generalized) in several ways, as for instance, causal BNs, dynamic BNs, Relational BNs, ...

In this lecture, we will focus on Bayesian network learning.
BN learning can differ with respect to the task : generative model versus discriminative one ? Then, the learning task can also differ w.r.t the nature of the data : complete data, incomplete data, non i.i.d data, number of variables >> number of samples, data stream, presence of prior knowledge ...

Given the diversity of these problems, many approaches have emerged in the literature. I will present a brief panorama of those algorithms and describe our current works in this field.

Abstract
Traditional user interface design works best for applications that only have a relatively small number of operations for the user to choose from. These applications achieve usability by maintaining a simple correspondence between user goals and interface elements such as menu items or icons. But we are entering an era of high-functionality applications, where there may be hundreds or thousands of possible operations. In contexts like mobile phones, even if each individual application is simple, the combined functionality represented by the entire phone constitutes such a high-functionality command set. How are we going to manage the continued growth of high-functionality computing?
Artificial Intelligence promises some strategies for dealing with high-functionality situations. Interfaces can be oriented around the goals of the user rather than the features of the hardware or software. New interaction modalities like natural language, speech, gesture, vision,  and multi-modal interfaces can extend the interface vocabulary. End-user programming can rethink the computer as a collection of capabilities to be composed on the fly rather than a set of predefined "applications". We also need new ways of teaching users about what kind of capabilities are available to them, and how to interact in high-functionality situations.

Abstract

Since the 1990s, intelligent agents and multi-agent systems with their known characteristics given by the AI community have been used to intelligently support individual or groups of human users in their business or personal decision-making in a large number and variety of applications in the Internet and Web. The developed frameworks of and systems with agent-based decision support differ in the extent to which they cover the typical phases of human decision-making, the types of decisions to be made, and their effectiveness and efficiency of support. In this respect, recommender systems, automated negotiation systems, matchmaker agents, and virtual user agents or avatars are classical examples to name but a few. On the other hand, semantic web technologies have been used during the past decade for providing better decision support through means of formal ontology-based representation, discovery, integration, and sharing of relevant heterogeneous data, knowledge, and services. Both agents and semantic web have deep and common roots in AI. However, despite the tremendous advances in research and development made in each of both technology domains, the full synergetic potential of agent-based and semantics-empowered decision support in innovative applications in the future Internet remains to be unlocked.

In this talk, I will present and discuss the basic principles, potential and challenges of a single and combined use of agents and semantic technologies for intelligent decision support together with selected showcases from different application domains

Abstract

Recent research has demonstrated the importance of cognition, motivation and especially of the emotion in the learning process. The link between emotion and learning appears fundamental and to measure the emotional behaviour several physiological devices have been used  such as seats positions, skin conductivity, face reading, sensitive mouse, blood pressure, or even cardio rhythms. Usually, these sensors measure physical reactions and external appearance of the learner.

New environments have emerged with new sensors such as electroencephalogram (EEG). They allow to save the brain activity of the learners during their learning tasks and to detect their emotions. This electrical activity can play an important role not only in the detection of brain states but also to help the learner to be in the best conditions of learning. The Brain itself can then be considered as an important agent. In   measuring mental activity we could induce pedagogical advices or adequate emotional environments to strengthen learning. We will illustrate how these components open new perspectives in various domains

Abstract
In the last decades, for many people computer games have replaced books and movies as their main source of entertainment. In games, players interact with virtual characters, which are often controlled by artificial intelligence. As these virtual characters act autonomously in their virtual environment, they can be thought of as situated agents. A big challenge in controlling the behavior of these agents is that many games place heavy requirements on the use of resources and the expression of the agents' behaviors. The requirement that might be the hardest to meet, is that the agents' behavior should appear "natural" within the virtual environment. 

For commercial video games, developers often employ techniques that circumvent the autonomy of the agents ("cheating"). This may be considered acceptable from the perspective of entertainment. However, in so-called "serious games" (games for training and education), the shortcuts that commercial developers tend to use may defeat the game's purpose. Therefore, for serious games, tools and techniques that are more advanced than the commonly used ones, are a necessity.

In this talk, I will give an overview of the state-of-the-art of the implementation of virtual agents in games, and compare this with the requirements that these agents must meet. I will then discuss some of the techniques developed by academic researchers to implement the behavior of game agents, in particular concerning their ability to adapt to the dynamics of the environment, and their ability to model the behavior of other agents, including those controlled by humans.

Abstract

Big Data is often used to denote a collection of data that is too big to process by conventional data analysis. In many fields there is an abundance of data, for example in chess, particle physics, and astrophysics. This keynote aims to shed light on this emerging field, to show present and future technical challenges for artificial intelligence and, above all, to demystify big data from meanings and opinions that are given to them by the laymen.

Almost all available knowledge is currently stored in large databases. Traditional data extraction methods assume that the data under investigation is structured in such a way that it is already known what the relevant information is. For Big Data, it is neither known in advance what is relevant, nor how to extract the relevant data from the data banks. Thus one of the central challenges is how to identify and extract relevant data from this enormous data source.

A second crucial difference between traditional data processing and the processing of big data is their focus. The focus has shifted from establishing causal relations to detecting correlations, In practice, correlations are easy to derive from data. Moreover, they form the relevant information for users and customers. As a case in point, we mention a well-known result in crime prevention. It is found by correlations that when it is warm outside, more crimes happen around supermarkets. The causal relation between the two is not immediately obvious, but does not affect the law-enforcement policy: there should be more police around supermarkets if it is hot, regardless of the cause of the crime. The observation that correlations are replacing causal relations in many domains, is called the computational turn. Effectively, this characterize the importance of Big Data. Moreover, finding correlations requires little intelligence, artificial or otherwise.

It follows that one of the challenges of Big Data lies in the interpretation of (correlational) results: by (1) visualization of data and (2) the forming of a narrative. Visualizing the data is important to understand the correlations that are found. It is also the first step of a narrative: what story does the data tell? For example, the Google Flu database shows regional prevalence of the flu. A narrative would be to find out the underlying cause as to why this is the case.

Big Data is a promising area for future research in artificial intelligence. First of all, the identification of relevant data is a challenging task and will continue to be more challenging as the data collected grows exponentially. Next, there is a theoretical challenge: currently, the computational turn has shifted the focus to correlation. However, as the intelligence of computer programs progresses, it may become possible to reason about data and make models and derive causations from correlations again.

Below we mention ten applications, of which two will be discussed during the keynote lecture.

1. Safety (politics, military)
2. Public Safety (Live View)                
3. Commerce (ads)
4. Banking (money streams)
5. Judiciary (CODR)
6. Waterway transport                         
7. Communication (twitter, phablet)
8. Education (MOOC)
9. Public governance
10. Warfare (Multi Agent Systems, Socio Cognitive Models)

To conclude, if advances in AI are able to challenge the causation problem, the Big Data of today becomes the traditional data of tomorrow.