Rat Maze Applet

This applet is an illustration of the use of the Neurosolver, a neural network based on a cortical column, as a brain of a hungry rat running in a maze in a quest for food.

Guideliness on use

• Build a maze by clicking on the walls. Position rat or food by dragging it to a desired location.
• Run the rat by pressing the < Run > button (the button changes to < Pause >). This action activates a drive in the rat's limbic system.
• Select the preferred running strategy from a number of choices:
• < random > enforces a completely random selection of rat's movements,
• in < constrained > mode, the rat chooses moves randomly, but it discards backward moves,
• < depth-first > search guides the rat in such a way as to visit each location only once; then, the cycle is repeated,
• selecting < Neurosolver > forces the rat's brain to solve the task; if a solution canot be found, then the mode switches to < constrained > and learning is activated.
• The rat will stop running if it eats all of the food or if the < Pause > button is clicked (the button changes to < Run >).
• The rat can explore the maze without stopping by hiding all of the food by clicking on < No food >.
• Place another piece of food by clicking on < More food >. Drag it to the desired location, then request another piece. Placing food corresponds to activating of imaginary (rat's memory trace of the food) or sensory perception.
• Check < Train Neurosolver > if you want the rat to learn.
• Rat can be taught automatically by letting it run using any strategy or manually by dragging it. If learning, the rat cannot be dragged through the walls.
• If the rat cannot move using the < Neurosolver > (due to insufficient knowledge or because of the modifications to the maze), then depending on the settings to < Inhibition: > one the following will happen:
• if there is no inhibition enabled (< none > selected), the strategy will switch to < constrained random >, learning will be enabled and the rat will continue the explorations; the rat will adapt by enforcing alternative routes;
• if < gradual > inhibition is selected, then the rat will gradually loose its memories about the paths that are not valid by inhibiting the faulty passages;
• if < immediate > inhibition is selected, then the memory about the problematic link in the path (a new wall) is erased at once.
• The memory (Neurosolver) of the rat can be controlled by selected proper choices in < Strength: > in the following ways:
• < uniform > does not use probabilities; the rat will always look for the shortest path (if using the Neurosolver for searches);
• < unidirectional > uses only succession probabilities; most common paths are followed first;
• < bidirectional > uses both, succession and precedence probabilities; even more pressure is put on selecting paths that were used most often in the past.
• Click on < Reset memory > to erase whatever the rat learned.
• Click on < Restart > to start from scratch.
• The speed of the rat is controlled by the number of < Steps: > per second (hardware constrained).
• The < Size: > of the maze can be changed by typing in a desired value and pressing the < Restart > button. This will restart the maze.

The maze is described in one of my papers.