TU Wien:Selbstorganisierende Systeme VU (Rauber)/Prüfung 03-03-2022
Zur Navigation springen
Zur Suche springen
Everything from memory:
20 True/False Questions, Time 20 Minutes, Correct Anser = 100%, Incorrect Answert = -50%:
- radius based neighbourhood graphs focus on cardinality over density
- in smoothed data historgams you can see cardinality and density clusters
- in smoothed data histograms you can see cluster borders
- Emergent SOMs rather show topological structure but ignore vector mapping.
- In SO discretization, methods that discretize the search space can't handle combinatorial solutions.
- In SOM visualization, Cluster Connection and D-Matrix are based on the same data.
- In Smoothed Data Histogram, reducing the smoothing factor leads to greater granularity in the cluster.
- CA is a gradient based.
- in metromaps stations stand for intersections between clusters
- fluctuations are for regulation and reinforcements for generating new solutions
- stigmergy does not apply to robotics
- ants use local and global information
- hybrid methods can be used to mitigate the poor convergence behaviour of PO
- CA lattice is 1D or 2D
- ......
4 Open Questions, Time 25 Minutes:
- What is intensification and diversification? How do they affect swarm behaviour and what parameters can be changed to effect them?
- How does SOM training work? What are the differences/advatages to Clustering? What is the complexety of SOM training?
- Describe 2 practical examples of CA (including lattice, neighbouhood, state transitions,...)
- Which properties of self organisation apply to ant algorithms and which do not and why?
- What do the PO paramters inerta, personal confidence, swarm confidence mean? How do they ralate to inensification and diversification?
- Differences between CA and Multi-Agent-Systems?
- How do pheromones in ACO work?