diff --git a/content/introduction.tex b/content/introduction.tex
index 1372204..50442c8 100644
--- a/content/introduction.tex
+++ b/content/introduction.tex
@@ -566,7 +566,32 @@ This work was presented at ICGI:
 \startsection
   [title=Conclusion and Outlook]
 
-\todo{Twee wegen. Komen ze nog bij elkaar? Nut van nominale technieken.}
+With the current tools for learning, it is possible to learn big state machines of black box systems.
+However, a real bottleneck is conformance checking of the hypothesis.
+This thesis provides background on conformance checking and also investigates a new algorithm.
+
+These testing algorithm are as efficient as they can.
+So in order to improve on this bottleneck, one possible direction is to consider \quotation{grey box testing}.
+This means that we should be looking into using more information of the system during testing (and learning).
+Often, we do have (parts of the) source code and we do know relationships between different inputs.
+The question is how this additional information can be integrated in the learning and testing of systems.
+
+Another path taken in this thesis is the research on nominal automata.
+This was motivated by the problem of learning automata over an infinite alphabet.
+So far, the results on nominal automata are mostly theoretical in nature.
+Nevertheless, we show that the nominal algorithms can be implemented and that the algorithms can be run concretely on black box systems (\in{Chapter}[chap:learning-nominal-automata]).
+
+However, the tools leave much to desired in terms of efficiency.
+Some of the efficiency is tackled in \in{Chapter}[chap:ordered-nominal-sets] for a particular symmetry.
+Another result is the fact that some automata can be \quotation{compressed} if they accept a certain type of language (\in{Chapter}[chap:separated-nominal-automata]).
+
+Last, it would be interesting to marry the two paths taken in this thesis.
+I am not aware of complete test suites for register automata or nominal automata.
+The results on learning nominal automata in \in{Chapter}[chap:learning-nominal-automata] show that this should be possible, as an observation table should give a test suite.
+However, there is an interesting twist to this problem.
+The test methods from \in{Chapter}[chap:test-methods] can all account for extra states.
+For nominal automata, we should be able to cope with extra states and extra registers.
+It will be interesting to see how the test suite grows as these two dimensions increase.
 
 
 \stopsection
diff --git a/environment.tex b/environment.tex
index be74996..380063f 100644
--- a/environment.tex
+++ b/environment.tex
@@ -67,7 +67,8 @@ This chapter is based on the following publication:
 
 \startsetups LogoSetup
 \setlayer[Logo]{
-\framed[frame=off, width=1cm, height=1cm, background=color, backgroundcolor=gray, foregroundcolor=white]{\getmarking[chapternumber]}
+\clip[width=1cm, height=1cm, hoffset=1.25cm, voffset=1.25cm]
+{\scale[factor=100]{\rotate[rotation=-20, frame=off, width=1cm, height=1cm, background=color, backgroundcolor=gray, foregroundcolor=white]{\pagenumber}}}
 }
 \stopsetups