In the field of evolutionary robotics artificial neural networks are often used to construct controllers for autonomous agents, because they have useful properties such as the ability to generalize or to be noise–tolerant. Since the process to evolve such controllers in the real– world is very time–consuming, one usually uses simulators to speed up the evolutionary process. By doing so a new problem arises: The con- trollers evolved in the simulator show not the same fitness as those in the real–world. A gap between the simulated and real environments exists. In order to alleviate this problem we introduce the concept of neuromodu- lators, which allows to evolve neural networks which can adjust not only the synaptic weights, but also the structure of the neural network by blocking and/or activating synapses or neurons. We apply this concept to a peg–pushing problem for KheperaTM and compare our method to a conventional one, which evolves directly the synaptic weights. Simula- tion and real experimental results show that the proposed approach is highly promising.
The basic idea for ANNs is taken from biology. | "Artificial brain" | weighted graph consisting of neurons (nodes) and connections (edges) | approximates solutions with large numbers of input.
creating and deleting connections weight changes of connections addition and removal of neurons.
NMs are substances that can dynamically influence several properties of neu- rons and therefore the function of a neural network. In contrast to neuro- transmitters(NTs) the effect of NMs spreads slower and lasts longer. NMs change the processing characteristics of neural networks by affecting the membrane po- tential, the rate of changing the synapses(i.e. influence on learning mechanisms) and other parameters. Typical NMs are acetylcholine, norepinephrine, serotonin, dopamin(all are also used as NTs), somatostatine and cholecystokinine(both also used as hormones in the human body) and many small proteins. Although these substances are released in a less local manner than NTs, the effects can be quite specific. This specificity comes from specific receptors on the neurons and their synapses.
Evolution is somehow difficult to define. But we think, that the University of Cal- ifornia, Berkeley defines it best. On their information website for evolution it says that ”Biological evolution, simply put, is descent with modification“. Furthermore, the authors describe the ”central idea of biological evolution is that all life on Earth shares a common ancestor“ and thus, that ”Evolution means that we’re all distant cousins: humans and oak trees, hummingbirds and whales“. In our case, evolution is conducted by SIMMA. The output of the evolutionary progress shows, that survival of the fittest is the main aspect in our simulation. After setting up parameters like population-size, evolution-duration, etc., SIMMA was able to evolve the best geno- type (ANN or NMN, depending on what we wanted to evolve in the first place), and then created a ’brain’, which we were able to load into SIMMA again and run experiments with it.
Presentation and Discussion in Class
Milestones, due dates and what I already achieved.
Goals for the future and deadlines.
November, 30th 2018
December, 4th 2018
December, 11th 2018
December, 15th 2018
December, 18th 2018
January, 7th 2019
due date
due date
My progress so far.
coming soon...
Coming Soon.
Report, final version.
Anmerkungen 1
Comment, oder so.
Anmerkungen 2
Comment, oder so.
Universität Salzburg | Fachbereich Computerwissenschaften
Jakob-Haringer-Strasse 2
5020 Salzburg
fbleitung@cosy.sbg.ac.at
+43 662 8044-6300