Department of Knowledge
Engineering

SIKS-DKE Colloquia

**
2014**

Title: Using evolutionary game theory to define, understand and treat cancer

Speaker: Prof. Dr.
Joel Brown, Department of
Biological Sciences, University of Illinois at Chicago.

When and Where:

Date: Friday,
**October 31**, 2014

Time: 13:00-14:00

Room: 0.015

Location: Maastricht University, Department of Knowledge Engineering,
Bouillonstraat 8-10

Abstract:
Metastatic cancer represents the evolution of a new, single-celled, asexual
protist. With the onset of
malignancy, cancer becomes a disease of Darwinian evolution as the tumor cells
mutate, evolve, adapt and speciate in response to their circumstances.
Tumors, represent “ecosystems” complete with ecological and evolutionary
dynamics. We can expect tumor cells
to multiply and evolve in response to the normal tissue, to physical
characteristics such as glucose, oxygen and Ph, and to other tumor cells.
Evolutionary game theory, a branch of applied mathematics that integrates
ecological and evolutionary dynamics, is well suited to modeling cancer
progression and treatment resistance.
We shall start with how and why cancer is an evolutionary game and move
onto how tumor ecology and evolution can be incorporated into cancer models.
In an example with breast cancer, data from biopsies and cell staining
techniques test and inform the models; showing cancer cells diversifying into
predictable ecological niches. Finally, evolutionary game theory can be used to
model treatment resistance – the moment therapy is initiated the physician has
entered into a predator-prey game with the tumor cells.
The goal of these models is to support evolutionarily enlightened therapy
by incorporating the known states of the cancer as well anticipating its
evolutionary responses to diverse treatment options. We shall conclude with an
example of such adaptive therapy with a class of disseminated prostate cancer.

Title: Competing Learning Heuristics in Cournot Games (joint with C.
Hommes and J. Tuinstra)

Speaker: Dr. **Marius Ochea**, UvA

When and Where:

Date: Thursday,
**October 16**, 2014

Time:16:00-17:00

Room: 2.015

Location: Maastricht University, Department of Knowledge Engineering,
Bouillonstraat 8-10

Abstract: We consider the
evolutionary competition between learning rules in Theocharis (1960) seminal
work on the stability of equilibrium in multi-player quantity-setting
oligopolies. The setting where firms may switch between heuristics based upon
past performance nests the famous Theocharis instability threshold, n=3, as a
special case. For competition between the Cournot and the rational heuristic,
both the existence and the magnitude of this threshold depend on the information
costs associated with the rational heuristic. When information costs are
positive a bifurcation route to chaos occurs as the number of firms increases.

Title:
Analyzing Phylogenetic Treespace

Speaker: Prof. Dr. Katherine St.
John, City University of New York

When and Where:

Date: Monday,
**September 22**, 2014

Time: 16:00

Room: 0.009

Location: Maastricht University, Department of Knowledge Engineering,
Bouillonstraat 8-10

Abstract: Evolutionary histories,
or phylogenies, form an integral part of much work in biology.
In addition to the intrinsic interest in the interrelationships between
species, phylogenies are used for drug design, multiple sequence alignment, and
even as evidence in a recent criminal trial.
A simple representation for a
phylogeny
is a rooted, binary tree, where
the leaves represent the species, and internal nodes represent their
hypothetical ancestors. For even
this simple way to represent evolution, finding the optima for a bimolecular
sequences for a fixed set of species is NP-hard.
This talk will focus on some of the elegant questions that arise from
improving search in this highly structured space.
We will also intriguing results about assembling, summarizing, and
visualizing the space of phylogenetic trees.
This talk assumes no background in biology and all are welcome.

Title: New insights into the dynamics of cardiac alternans and arrhythmias

Speaker: Prof. Dr. med.
Jan Kucera,
Department of Physiology, University of Bern, Switzerland

When and Where:

Date: Monday,
**September 22**, 2014

Time: 14:30

Room: 0.009

Location: Maastricht University, Department of Knowledge Engineering,
Bouillonstraat 8-10

Abstract: Cardiac alternans, i.e.,
the beat-to-beat alternation of the cardiac action potential duration (APD) is a
well-known mechanism that can cause severe heart rhythm disorders. Alternans can
result from dynamic instabilities of the interactions between membrane potential
and ion currents (voltage-driven alternans) or from instabilities of
intracellular calcium cycling (calcium-driven alternans). The propensity to
alternans is classically understood in terms of the function relating APD to the
previous diastolic interval. However, a number of studies showed that this
approach is not always reliable to predict alternans.

In contrast, theoretical studies demonstrated that the eigenvalue of the
dominant eigenmode represents an appropriate marker for the susceptibility to
alternans. Therefore, our aim was to develop a method allowing an experimental
estimation of this eigenvalue, and to examine further markers to discriminate
between voltage-driven and calcium-driven alternans. We developed a new
framework in which a cardiac cell or cardiac tissue is treated as a system
transforming an input into an output (e.g., pacing intervals into APDs). The
system is then characterized in the frequency domain by its transfer function
and its eigenvalues. For a practical implementation, a cardiac cell is paced at
intervals varying stochastically, the APDs are registered, and the poles and
zeros of the corresponding transfer function are identified.

This approach was tested and validated using computer simulations and
corresponding experiments with cardiac cells. The results confirm that this
approach can be used to predict the onset of alternans and to distinguish
between voltage-driven and calcium-driven mechanisms. This distinction is
important, because these two mechanisms may precipitate arrhythmias in different
manners. This work also underlines the importance of combining theory,
simulations and experiments to gain deeper insights into the mechanisms of heart
rhythm disorders.

Title: Decision Trees for Computer Go Features.

Speaker: Dr. **Steve Kroon**, Computer
Science Division, Stellenbosch University.

When and Where:

Date: Friday,
**June 27**, 2014

Time: 14:00-15:00

Room: 0.015

Location: Maastricht University, Department of Knowledge Engineering,
Bouillonstraat 8-10

Abstract: Monte-Carlo Tree Search
(MCTS) is currently the dominant algorithm in Computer Go. MCTS is an asymmetric
tree search technique employing stochastic simulations to evaluate leaves and
guide the search. Using features to further guide MCTS is a powerful approach to
improving performance. In Computer Go, these features are typically comprised of
a number of hand-crafted heuristics and a collection of patterns, with weights
for these features usually trained using data from high-level Go games.

This talk presents joint work with a recently completed Masters student
investigating the feasibility of using decision trees to generate features for
Computer Go.

Title: **Resource Competition on Integral
Polymatroids
**

Speaker: Dr.

When and Where:

Date: Wednesday,
**May 21**, 2014

Time: 16:00-17:00

Room: 0.015

Location: Maastricht University, Department of Knowledge Engineering,
Bouillonstraat 8-10

Abstract: We study competitive
resource allocation problems in which a set of players distribute their demands
integrally on a set of resources subject to player-specific submodular capacity
constraints. Each player has to pay for each unit demand a cost that is a
nondecreasing and convex function of the total allocation of that resource. This
general model of resource allocation generalizes both singleton congestion games
with integer-splittable demands and matroid congestion games with
player-specific costs. As our main result, we give an algorithm computing a pure
Nash equilibrium.

The proof rests on a structural result on the sensitivity of optimal
solutions minimizing some linear objective over an integral polymatroid base
polyhedron which is of independent interest.

Title: **"Make it slow to make it pattern"**

Speaker: Dr. **Christian Fleck,
**Wageningen University.

When and Where: **CANCELLED**

Date: Monday,
**May 19**, 2014

Time: 16:00-17:00

Room: 0.015

Location: Maastricht University, Department of Knowledge Engineering,
Bouillonstraat 8-10

Abstract: Turing pattern on cellular tissues are often described by a set of
coupled non-linear ordinary differential equations. To arrive to these equations
it is assumed that intra-cellular diffusion is orders of magnitude faster than
the transport across the cell membranes. This is often an oversimplification; in
particular macromolecular crowding can severely hamper intra-cellular diffusion.
Relaxing this simplifying assumption yields a hybrid system described by a set
of non-linear partial differential equations coupled via the boundary
conditions. We show that the additional degree of freedom on the microscopic
length scale yields pattern on the macroscopic scale where the corresponding
ordinary differential system would not exhibit any structure.

Title: **Dynamics of spatial pattern
networks in generalized rock-paper-scissors models**

Speaker: Dr. **Josinaldo Menezes**,
post-doc at Institute for Biodiversity and Ecosystem Dynamics (IBED) -
University of Amsterdam, and
Assistant Professor at University of Rio Grande do Norte - Natal, Brazil.

When and Where:

Date: Wednesday,
**May 14**, 2014

Time: 14:00-15:15

Room: 0.015

Location: Maastricht University, Department of Knowledge Engineering,
Bouillonstraat 8-10

Abstract: We study the population dynamics in generalized rock-paper-scissors
models with an arbitrary number of species. We perform stochastic and
deterministic numerical simulations and observe the formation of domains, where
individuals of one or more species coexist, separated by interfaces whose
dynamics is curvature driven. We investigate the formation of such patterns by
using scalar field theory and show that the coarsening of these networks follows
a scaling law which is analogous to that found in other physical systems in
condensed matter and cosmology. We also investigate formation of competing
partnerships and show that spiral waves can arise both for odd and even number
of species.

Title: **An overview of mCRL2: theory,
tools and applications**

Speaker**: **prof.dr.ir. **Jan-Friso Groote**, TU Eindhoven.

When and Where:

Date: Wednesday,
**30 April**, 2014

Time: 16:00-17:00

Room: 0.015

Location: Maastricht University, Department of Knowledge Engineering,
Bouillonstraat 8-10

Abstract: In 1990 we started to
develop an extension of process algebra to allow convenient modelling of and
reasoning about real systems. It quickly became obvious that we needed
assistance by tools, as most real systems are far too complex to prove correct
by hand. The result is a compact but very expressive language, which we call
mCRL2, to model behaviour, a strong underlying theory, a very expressive modal
logic to denote properties, and a supporting toolset which allows to model,
reduce, verify and visualise behaviour. See
www.mcrl2.org.

The language has been for many kinds of applications, for instance at CERN in
Geneva, where it was used to prove certain aspects correct of networks of many
thousands of finite state machines that govern the experiments.

In the summer of 2014 the book "Modelling and Analysing of Communicating
Systems"

will appear at MIT press describing especially the theory behind mCRL2.

Title: **Evolutionary Environmental Games**.

Speaker: Dr.** ****Paolo Zeppini**,
Section Mathematical Economics & Mathematics, UvA,

When and Where:

Date: Wednesday,
**April 9**, 2014

Time: 16:00-17:00

Room: 0.015

Location: Maastricht University, Department of Knowledge Engineering,
Bouillonstraat 8-10

Abstract: We propose a model for the strategic interaction of a large number of
players in environmental public goods such as lakes, river basins, woodlands, or
crop plantations. We specifically consider complex ecological systems with a
non-linear response to human pollutant activities. The starting point of
analysis is the economy of shallow-lakes (Brock and de Zeeuw, 2002). Instead of
the traditional game theoretic setting of interaction between two or few
perfectly rational strategic players, we adopt an evolutionary approach with a
large population of myopic players. Each player represents an economic agent
with an incentive to exploit the environmental public good. In the case of a
lake, agents can be farmers and firms, but also families and tourists. In all
cases the public good problem builds on the trade-off between the quality of the
"environmental" services provided by the ecological system. The nature of the
2x2 stage interaction game changes with the state of the ecosystem. We are able
to classify different conditions for stable clean regimes, stable polluted
regimes, stable regimes with co-existence of cooperators and free-raiders,
weakly stable equilibria, and dynamic regimes. In particular dynamic regimes
present a rich variety of long-run attractor, including chaotic dynamics.

Title: **Multiplayer Cost Games with
Simple Nash Equilibria**

Speaker: Dr.** Julie de Pril, **Département de Mathématique, Université de
Mons, Belgium

When and Where:

Date: Tuesday
**January 28**, 2014

Time: 12:30-13:30

Room: 0.015

Location: Maastricht University, Department of Knowledge Engineering,
Bouillonstraat 8-10

Abstract: Computer aided verification, and more specifically model-checking, is
a branch of computer science which offers techniques to check automatically that
a given computer system satisfies a given specification. The classical
model-checking techniques apply on a model or an abstraction of the system (such
as a finite automaton) together with a translation of the specification into a
logical formula (of a temporal logic for instance).

In this talk, we will see that recent research has evolved to obtain adequate
models for computer systems that become more and more complex and to measure the
adequacy of a specification (in term of quality of service, time and/or energy
consumption,...). In particular, we resort to multiplayer non zero-sum
quantitative games played on graphs (also called multiplayer cost games).

We show that a large class of these games, including games where the individual
objectives are mean- or discounted-payoff, or quantitative reachability, does
not only have a solution, but a simple solution. More precisely, we establish
the existence of simple Nash equilibria. Realistically, agents have to decide on
their moves with very limited resources, and complicated strategies that require
exponential -- or even non-elementary -- memory cannot realistically be
implemented. The existence of simple strategies therefore holds a promise of
implementability.

Title: **On the Existence of Approximated
Equilibria and Sharing-Rule Equilibria in Discontinuous Games (joint work with
Philippe Bich) **

Speaker: **Rida Laraki, **Directeur de Recherche at CNRS in LAMSADE and
professor at Ecole Polytechnique, Paris

When and Where: (joint ETBC seminar)

Date: Thursday,
**January 16**, 2014

Time: 16:00-17:30

Room: A1.22 at TS 53, School of Business and Economics, Maastricht
University

Abstract: New relaxations of the Nash equilibrium concept are shown to exist in
any strategic game with discontinuous payoff functions. The new concepts are
used (1) to show the equivalence between Reny's better-reply security condition
[1] and Simon-Zame's endogenous tie-breaking rule equilibrium concept [2], (2)
to obtain conditions for the existence of approximated equilibria in a class of
discontinuous games that naturally extends Reny's better-reply secure games, and
(3) to show the existence of approximated equilibria in a large family of
two-player games that contains all standard models of auctions.

[1] Reny P.J. (1999). On the Existence of Pure and Mixed Strategy Nash
Equilibria in Discontinuous Games. Econometrica, 67(5), 1029-1056.

[2] Simon L.K. and Zame W.R. (1990). Discontinuous Games and Endogenous Sharing
Rules. Econometrica, 58, 861-872.

Title: **True Online TD(λ)**

Speaker: Dr. **Harm van Seijen**, Department of Computing Science, University
of Alberta

When and Where:

Date: Wednesday,
**January 8**, 2014

Time: 14:00-15:00

Room: 2.015

Location: Maastricht University, Department of Knowledge Engineering,
Bouillonstraat 8-10

Abstract: TD(λ) is a core algorithm of modern reinforcement learning. The appeal
of TD(λ) comes from its clear and conceptually simple forward view, and the fact
that it can be implemented online in an inexpensive manner. While offline TD(λ)
matches the forward view exactly, online TD(λ) --- which is more interesting for
practical applications --- only approximates it. Up to now, it has been an open
question whether a version of online TD(λ) could be made that matches the
forward view exactly. We introduce a new online TD(λ) algorithm for function
approximation, with the same complexity as the regular version, that achieves
the forward view exactly. Key to this is a refinement of the online version of
the forward view that is well defined at each time step rather than only at the
end of an episode. We use this refined version to derive the new algorithm. By
adhering more truly to the goal of matching the forward view, the new algorithm
performs much better in practise. We demonstrate this on several standard
benchmark problems, where it outperforms both accumulating and replacing traces.