The Interactive Intelligence group of Delft University of Technology is one of the top groups in the world in the area of automated bilateral negotiation strategies. In automated negotiation software agents are negotiating against other agents on behalf of humans, but without the intervention of humans. These automated negotiation agents outperform humans in the sense that deals made by the agents are significantly better than those made by humans. However, there are a number of constraints that have to be met before the automated negotiation agents can do their work:

·         The negotiating agent expects a preference profile to be given to it, and preference elicitation is notoriously difficult.

·         The technology is still largely unknown to the public and it is by no means certain that people will trust the technology enough to let agents negotiate on their behalf.

To address these issues and in cooperation with expert negotiators in The Netherlands we are developing negotiation support systems that make use of the best automated negotiation strategies to advice human negotiators. In an ongoing project we have developed prototypes for the non-expert negotiator that include preference elicitation tools. In this project we want to create a version that is dedicated to the expert negotiator.

The PhD student will research and develop negotiation strategies that can find near optimal solutions in only a few rounds of negotiation. This is a challenge as human negotiations take between 3 and 30 rounds (depending on the national culture of the negotiators), and not the thousands of rounds that are typical for automated negotiations. The PhD student will take existing elements of the negotiation support systems already developed in the group to build a prototype negotiation support for experts and add into this the negotiation strategies he/she developed. The prototype is then tested by the PhD student and improved in cooperation with negotiation experts.

Candidates for this project should have a master degree in computer science, and an interest in game theory, multi-issue decision making, and supporting humans.

More information on the negotiation projects going on in the group and the negotiation systems we developed can be found here:

·         Prof. dr. C.M. Jonker (c.m.jonker@tudelft.nl, ii.tudelft.nl/)

·         Jeroen Meijer, solicitor/lawyer and member of the board of the Netherlands Institute of Negotiation (http://www.stichtingnio.nl/)

 

Intelligent software and robots are gaining ground in society. Apps and software agents of all kinds are to help humans in their daily business, and soon robots will drive our cars for us. Interactivity with other intelligent entities (humans, robots and apps/agents) is essential for their success. The Interactive Intelligence section of Delft University of Technology is developing the next generation of such intelligent robots and agents that is to be characterised by their ability to form a Theory of Mind of the intelligent entities they have to interact with and work with. In this project proposal the focus is on developing computational models of intelligence (also called cognitive models) to be used in robots and agents that are capable of strategic thought and capable of reflecting on the strategic thoughts of other robots or agents.

The project will be tested in the environment of the Blocks World for Teams (BW4T) that allows for humans, agents and real robots to work together to perform a team task.

The PhD student will research and develop strategic models for game play in the BW4T environment, the cognitive models (theory of mind) that allow the intelligent agents and robots to reason about the strategies and minds of other players in the game and include these reflective thoughts in their own strategic reasoning. Experiments in simulation mode and with humans will be held to test the effectiveness and computational complexity of the strategic models and cognitive models. The models will then be deployed in robots (such as Nao robots) and tested in teams consisting of robots only and in teams consisting of robots and humans.

Prof. dr. C.M. Jonker (c.m.jonker@tudelft.nl, ii.tudelft.nl/). Prof Jonker is head of the Interactive Intelligence section of Delft University of Technology and is an expert in agent technology, mental models and strategic reasoning.

·         Prof. dr. L.C. Verbrugge (L.C.Verbrugge@ai.rug.nl, http://www.rinekeverbrugge.nl/), head of Multi-Agent Systems group of the University of Groningen.

Theory of mind is the ability to attribute mental states—beliefs, intents, desires, pretending, knowledge, etc.—to oneself and others and to understand that others have beliefs, desires, and intentions that are different from one's own. See e.g., Horst (2011).

B. Dunin-Kęplicz and R. Verbrugge, A logical view on teamwork. In: J. van Eijck and R. Verbrugge (eds.),Games, Actions and Social Software: Multidisciplinary Aspects, Texts in Logic and Games,FoLLI subseries of Springer Lecture Notes in Computer Science, volume 7010, Springer Verlag, Berlin, 2012, pp. 184-212.

Horst, Steven, "The Computational Theory of Mind", The Stanford Encyclopedia of Philosophy (Spring 2011 Edition), Edward N. Zalta (ed.), URL = <http://plato.stanford.edu/archives/spr2011/entries/computational-mind/>.

Harbers, M., Riemsdijk, M.B. van, & Jonker, C.M. (2012). Measuring Sharedness of Mental Models and its Relation to Team Performance. In Proceedings of The 14th International Workshop on Coordination, Organisations, Institutions and Norms, (COIN@AAMAS 2012).

Jonker, C.M., Riemsdijk, M.B. van, Kieft, I. van de, & Gini, M. (2012). Towards Measuring Sharedness of Team Mental Models by Compositional Means. In Proceedings of the 25th International Conference on Industrial, Engineering and Other Applications of Applied Intelligent Systems (IEA/AIE 2012).

B. Meijering, H. van Rijn, N.A. Taatgen, and R. Verbrugge, I do know what you think I think: Second-order theory of mind in strategic games is not that difficult. In: L. Carlson, C. Hölscher, and T. Shipley (eds.), Proceedings of the 33rd Annual Conference of the Cognitive Science Society, Cognitive Science Society, Austin, TX, 2011, pp. 2486-2491.

Rens, T., Jonker, C.M., Riemsdijk, M.B. van, & Wang, Z. (2012). A Multi-Party Negotiation Game for Improving Crisis Management Decision Making. In Proceedings of the 25th International Conference on Industrial, Engineering and Other Applications of Applied Intelligent Systems (IEA/AIE 2012).

R. Verbrugge, Logic and social cognition: The facts matter, and so do computational models. Journal of Philosophical Logic 38 (6), 2009, p. 649-680.

H. de Weerd, R. Verbrugge and B. Verheij, Higher-order social cognition in rock-paper-scissors: A simulation study. In: G. Bonanno, H. van Ditmarsch and W. van der Hoek (eds.), Proceedings 10^th Conference on Logic and the Foundations of Game and Decision Theory (LOFT 2012), University of Sevilla, 2012, pp. 218-232.

 

 

Digital visual media are nowadays the principal support for information exchange and consumption. Their amount is exponentially growing and needs to be organized and delivered to the final user in an effective way. In doing this, a major challenge is to optimize the media content so that it fulfills technology limitations (such as bandwidth constraints or imperfections in imaging devices), yet meeting the user’s expectations in terms of visual quality. So far content optimization has been based on algorithms that estimate visual quality from measures of signal fidelity, lately integrated with models of visual perception. Promising results obtained in a preliminary phase, but are currently coming to a stalling phase. One possible reason for this is that current algorithms lack in modeling the dependency of visual quality preferences on cognitive mechanisms working on top of perception (e.g., engagement with the content), which has been recently proved to exist [1].

This PhD project is meant to move the first steps towards the integration of high-level cognitive information into automated visual quality assessment. The candidate will focus on studying the added value of semantic information into visual quality assessment methods. In particular, the project will focus on understanding the role that semantic categorization (i.e., our ability to assign “labels” to objects, such as “fruit” or “tool”) plays in visual quality preferences. As a first step, an intensive empirical research activity will be performed. Subjective experiments will be carried out on a large scale to verify whether semantic categories exist for which visual quality is more critical. In a second step, this information will be integrated with existing knowledge on quality perception to design a quality assessment system that can also take into account semantic information. To do so, the use of computer vision and machine learning is to be envisioned [2].

Prof. dr. Ingrid Heynderickx is professor at the Interactive Intelligence section of Delft University of Technology and is an expert in visual perception of display and lighting systems.

Dr. Judith Redi is an assistant professor at the Interactive Intelligence section of Delft University of Technology and is an expert in modelling image quality and visual attention.

P. Kortum and M. Sullivan, “The Effect of Content Desirability on Subjective Video Quality Ratings”. Human Factors: The Journal of the Human Factors and Ergonomics Society, 2010

 

 

The project Care@Home is a part of the European Union Ambient Assistant Living (AAL) Joint Programme that is a joint research and development-funding programme implemented by 11 partners in 4 European countries. The goal of the project is to deliver connected ICT-based assistive living solutions for the elderly in their homes via a smartTV. The smartTV is acting as a user-centered ‘hub’ providing reciprocal communication that connects elderly to their formal care network, nursery home facilitators, family, friends and communities, and services, including household, healthcare, exercise, social, entertainment and security. The technological solutions are aimed at enabling the elderly to live self-reliant and independently and to become or to remain active participants in the society. The Care@Home infrastructure supports communication for home including, personal sensors, devices and services applied on different platforms, i.e. smartTVs, PCs, tablets, and smartphones. Research and development activities within the Care@Home project involve multidisciplinary fields, e.g. medical science, psychology, design, and computer science, and are oriented toward the end-product involving collaborations with industrial partners within the consortium and all stakeholders as the end-users of the developed product.

Previous study has been done in designing user interfaces dedicated for elderly and an ongoing project has developed the architecture of Care@Home. Taking into account the previous results, the PhD student will research and develop intelligent components within Care@Home. The new smart Care@Home should be able to autonomously adjust the user display and available system actions to current goals, context and profile of the users. The context information should include the user’s current situation, wellbeing, home situation, and interaction performance, and should be assessed by interpreting both verbal and nonverbal aspects of user inputs and information coming from available sensors and devices applied in the Care@Home infrastructure. The PhD project includes evaluation of the developed smart Care@Home by involving elderly end-users, to address the usability issues and to assess proposed user-interaction concepts.

Candidates for this project should have a master degree in computer science, and specializations in both human-computer interaction and artificial intelligence fields are strongly preferred.

Dr. Nick Guldemond PhD DSc (med) (n.a.guldemond@tudelft.nl) is assistant professor at the Interactive Intelligence section of Delft University of Technology and is an expert in ICT solutions for medical care.

Dr. Ir. Siska Fitrianie, S.T., PDEng (s.fitrianie@tudelft.nl) is researcher at the Interactive Intelligence section of Delft University of Technology

http://www.careathome-project.eu/

 

A picture is flat (2D), yet you perceive depth (3D). Depth is added by the mind and we study this process. We use sophisticated experimental paradigms to measure the geometrical structure of pictorial space (perceived depth). For example, we programmed an experiment where the observer can adjust a 3D arrow on a surface such that the arrow points in the normal direction of that surface. In mathematical terms, this is equivalent to measuring the perceived (subjective) normal vector of the surface, which essentially is the derivative. If we integrate the normal vector over the whole surface, we get a perceived depth profile.

This is a way to measure the subjective depth of surfaces, of individual objects, but recently we became interested in the depth structure of the whole scene. The first method we developed (Wijntjes and Pont, 2010) was based on a pointing task. Observers are shown a 3D pointer in the pictorial scene and use that to point to a target location. Importantly, not only can the pointer be manipulated in the picture plane, but also in the depth direction. We developed an algorithm to reconstruct the overall depth of such settings and started analysing, for example, whether depth perception becomes better when you use a 3D TV, or a conventional display (the answer is that it depend on the content of the picture). Later, other scientists from our lab developed more methods (Wagemans et al., 2011b; Van Doorn et al., 2011; Wagemans et al., 2011a).

The first major problem is the reconstruction algorithm for the pointing task. Currently this is either done with linear perspective (Wijntjes and Pont, 2010) or orthographic perspective (Wagemans et al., 2011a) and it is unknown which method is actually best describing depth perception. Secondly, it is unknown how different probing methods can be combined. Measuring depth via a pointing task gives in some cases different results than using a relative size task. This sounds problematic but could also be used to get rich insights in how observers perceive depth. A third problem is the relation between global depth and perceived layout. The layout of a scene is a view independent property, while depth depends on the viewing position. There is not much known about this relationship. The perceived layout can be measured by using Multi Dimensional Scaling techniques.

Lastly, an optional subproject concerns pictorial relief. Reconstructing a surface on the basis of normal vectors requires the use of a pseudo-inverse matrix which costs quite a bit of computation time. Recently we started to develop applications in which consumers can make a 3D texture mapping of their pictures, in which the 3D shape is acquired via our experimental paradigm. In order to get this done in a reasonable amount of time, we need to optimise the pseudo-inverse methods. Since we have a very sparse matrix, it might be possible to solve this.

Dr. Maarten Wijntjes (www.maartenwijntjes.nl) is assistant professor at the Human Information Communication Design group of Delft University of Technology and is an expert in depth perception.

A.J. Van Doorn, J.J. Koenderink, and J. Wagemans. Rank order scaling of pictorial depth. i-Perception, 2(7):724–744, 2011.

J. Wagemans, A.J. Van Doorn, and J.J. Koenderink. Measuring 3d point configurations in pictorial space. i-Perception, 2(77-111), 2011a.

J. Wagemans, A.J. Van Doorn, and J.J. Koenderink. Pictorial depth probed through relative sizes. i-Perception, 2:992–1013, 2011b.

M.W.A. Wijntjes and S.C. Pont. Pointing in pictorial space: Quantifying the perceived relative depth structure in mono and stereo images of natural scenes. Transactions on Applied Perception, 7(4), 2010.