Brest State Technical University, Course "Contemporary Intelligent Information Technology (CIIT)"
From Students' Works (2013)
All One Problem (with 1000 binary genes, assuming 1 is good gene while 0 is not.
Each population includes 100 such chromosomes)
With 1 and 0 both being 50% ... By Ilya Mokin (2013).
- With 1 being only 0.1% ... By Evhuh Alexandr (2013).
=> Quite amazingly it evolves to all "1" chromosome though it takes longer generation.
Function Minimization Problem => serch for the point of minimum value of y
- 20-dimensional Sphere Model: y = (x1)^2 + (x2)^2 + ... + (x20)^2
- 2-dimensional Sphere Model: y = x^2 ... By Galina-Bezabrazova
=> It seemed easy to find a minimum, that is (0,0). The algorithm found it at 10-th generation.
- The other 2-D function: y = x^4 - 5 x^3 - 6 x^2 + 8x + 15
showing points on the graph along an evolution ... By Kalybski Vasili
Training Neural Networks
- Feedforward NN with 5-5-1 architecture to solve 5-even-parity Problem
- Feedforward NN with 5-5-1 architecture to recognize Palyndrome
A navigation in a gridworld - find a lucky dog!
- Starting from the center of the world searching for a sausage without any obstacles like wall
(Left) Routes of 6 dogs at the 1st generation ... By Boiko Svetlana
(Center) Routes of a lucky dog and un unlucky one at the beginning ... By Danilchuk Raman
(Right) Routes of the cleverest dog at the final generation - from Home to Sausage ... By Vasili Brutsky
(=> This looks like a graph of evolution but this is not. This is a trace of luckiest dog from home to sausage.)
Dog searches for a sausage in the world with wall: evolution & final route of the luckiest dog ... By Lutich Pavel
If the field is T-like in which sausage is center of the left end of the top regeon
(Left) The trace of the luckiest dog after (...) generation ... By Leonchuk Ksenia
(Right) Also unlucky dog at the 1st generation and so-so-lucky dog during an evolution are shown ... By Vladimir Romanyuk
A dog starts at the left end of the bottom of our gridworld looking for a sausage at the right end of the bottom.
In between the dog and sausage we have a dengerous cliff at the bottom.
If the dog falls down to the cliff dog will die.
The luckiest dog found in the last generation ... By Galina-Bezabrazova
Let's a dog make a navigation in a gridworld avoinding a pond
Navigation by 3-3-2 neural network
(Left) (i) Unlucky dog and (ii) not-so-but-still unlucky dog example ... By LIsenkov Roman
(Left) An example of lucky dog ... By Lohnitskiy Alexey
(Right) Another example of lucky dog in a different geography ... By Alexander Mokin
Traveling Salesperson Problem (TSP)
In TSP with 5 cities, e.g., we can calculate the tour of minimum length by hand ... By Krivitski Andrei
=> A little unusual but he create distance matrix at random instead of creating cities at random.
=> (Left) Distance matrix of a random creation of 5 cities; (Center ) The length of all possible (5-1)!/2=12 cities; (Right) GA converges to the minimum tour of 138.8
Yet another minimum length tour of 5 cities TSP ... By Popko Andrey
Assuming the starting city Brest at (0,0), 4 cities to be visited are created at random as Minsk (10,5), Pinsk (7,9), Kobrin (5,4), Prygani (3,2)
In TSP with 5 cities, e.g., we can calculate the tour of minimum length by hand ... By Evhuh Alexandr
=> (Left) The shortest route in each generation, which suggest the search is somehow random and seems a lucky success
In TSP with 5 cities, e.g., we can calculate the tour of minimum length by hand ... By Boiko Svetlana
=> In this case the evolution is rather normal but the route might be not smallest, i.e., trapped to a local minimum
A challenge to the TSP with 100 cities. ... By Danilchyk Roman
=> Does the tour look like minimum?