Example

This example deals with a representation of animal behavior and learning. The agent or artificial animal is generated and attempts to cope with the features of the world with its limited knowledge. The objective is to adapt to the problem presented to it through learning, gradually modifying its behavior related to movement in the space (Figure 3).

The agent used in the example carries an ANN (three layers: input layer where data enter the networks, hidden layer where they are classified, and output layer where decisions are made) that must learn by reinforcement (punishment and reward) the best strategy to get as much profit as possible from the world in terms of food. In order to catch its food, it must be in the same pixel as its prey (prey do not move). Unfortunately as all animals, it is not perfect. It has limited knowledge and capacities. What it can do is make a decision each unit of time upon three possibilities: keep moving straight, turn slightly to the left, or turn slightly to the right. These decisions are relative to its current direction.

As input it has only memory consisting of knowledge as well as knowledge of failure of the last three decisions. It lacks any sensorial capacity or knowledge ofits location in the space.

The toroidal world is a square area on the top of the screen (Figure 3); below it, a performance histogram will appear (Figure 4), describing by time intervals the catch that this predator has achieved through the actual animation run.

* Demonstration program will be available through the first author by an email request.

Figure 1 The Fish-based Decision Support System (FIDESS). Quantitative data can be modified by moving the sliders, while the classification results (shown in the lower-left part of dialog) change in real time. The very user-friendly GUI played a fundamental role in the acceptance of the method among ecologists and fish biologists who were not familiar with the underlying computational methods but are used to apply simpler biotic indices.

Figure 1 The Fish-based Decision Support System (FIDESS). Quantitative data can be modified by moving the sliders, while the classification results (shown in the lower-left part of dialog) change in real time. The very user-friendly GUI played a fundamental role in the acceptance of the method among ecologists and fish biologists who were not familiar with the underlying computational methods but are used to apply simpler biotic indices.

FIDESS

FIDESS

Figure 2 The GUI of FIDESS makes the underlying ANN completely transparent to the user. Quantitative data are entered in input fields, while binary data are entered by means of check boxes. Sliders are also available for quantitative variables, thus providing immediate feedback to the user, who can easily compare FIDESS actual responses with its expected behavior. Both fuzzy and crisp (i.e., defuzzified) classifications are provided in output.

Figure 2 The GUI of FIDESS makes the underlying ANN completely transparent to the user. Quantitative data are entered in input fields, while binary data are entered by means of check boxes. Sliders are also available for quantitative variables, thus providing immediate feedback to the user, who can easily compare FIDESS actual responses with its expected behavior. Both fuzzy and crisp (i.e., defuzzified) classifications are provided in output.

No matter which decision this agent takes, there is a possibility of failure, and no matter how fitted it is, there is a possibility of being in a deserted area. With its limited information, it cannot avoid being starved at some intervals, but as learning proceeds it changes its behavior, starting from a random walk and finishing with sinusoidal movement when it has encountered prey or with a more straightforward movement when it has not. This improves o o°o

Figure 3 Schematic representation of the toroidal world with patches of food and an animal track.

the chances of catching prey when it is close to a patch (Figure 4). This type of strategy has been described in predators when prey is found in clusters. A sinusoidal movement in a situation close to a patch will increase the probability ofkeeping close to the patch and a straight movement in the opposite case will increase the probability of escaping from an empty area. This behavior has been described from small predators like insects all the way to humans (fishing vessels).

This example not only shows the main feature of an ANN, its learning capability, but also a way of classifying situations and relating them in this case to actions. This artificial animal is forecasting and taking the best action course that will lead it to its prey.

Ecological informatics methods have great potential and are not limited to forecasting or classifying. In this example not only was this achieved, but also a representation of learning or animal behavior, and an IBM was developed at the same time.

Oplan Termites

Oplan Termites

You Might Start Missing Your Termites After Kickin'em Out. After All, They Have Been Your Roommates For Quite A While. Enraged With How The Termites Have Eaten Up Your Antique Furniture? Can't Wait To Have Them Exterminated Completely From The Face Of The Earth? Fret Not. We Will Tell You How To Get Rid Of Them From Your House At Least. If Not From The Face The Earth.

Get My Free Ebook


Post a comment