Sally's Weblog

Welcome!

sallypmangaba — Sat, 01 Mar 2008 08:15:16 +0000

Neural Network…do you have any idea about it?

This blog contains some of my articles about what I have learned about the Neural Network.

The articles will be mostly compost of my own examples for some specific topics.

This is a blog wherein you can have my personal articles about the Neural Network.

You can get some insights here from me about some information related to the topic mentioned above.

Neural Network

sallypmangaba — Sat, 01 Mar 2008 08:11:57 +0000

What are neural nets?

A neural net is an artificial representation of the human brain that tries to simulate its learning process. The term “artificial” means that neural nets are implemented in computer programs that are able to handle the large number of necessary calculations during the learning process. To show where neural nets have their origin, let’s have a look at the biological model: the human brain.

TYPES OF LEARNING

sallypmangaba — Sat, 01 Mar 2008 08:07:56 +0000

*Note: The following are my own examples for each type of learning.

1. Supervised- A neural net is said to learn supervised, if the desired output is already known.

Training a parrot to say “hello”

As an input the trainer will say the word “hello” repeatedly. The hidden layer is how the parrot absorbs what it is hearing from the trainer. The output is that the parrot will be able to say properly the word “hello”.

2. Unsupervised- Neural nets that learn unsupervised have no such target outputs. It can’t be determined what the result of the learning process will look like. During the learning process, the units (weight values) of such a neural net are “arranged” inside a certain range, depending on given input values. The goal is to group similar units close together in certain areas of the value range. This effect can be used efficiently for pattern classification purposes.

Training the dog to solve 1+1

In unsupervised learning, we already know the output. In this situation we know that the output is 2. The dog should bark twice. The trainer should let the dog adjusts its number of bark, because in unsupervised learning, inputs are being corrected.

3. Reinforced – interacting with an environment, an agent learns from the consequences of its actions, rather than from being explicitly taught and it selects its actions on basis of its past experiences and also by new choices. Training the dog to play in an obstacle course

In reinforcement learning, what is wrong and right is unknown. The dog is not aware what the right path would be. The dog will generate an action and observe its environment. The dog then, will select the proper action and path to reach the finish line, because in reinforcement learning there is a sequential decision – making task. If the dog reached the finish line, there will be a reward, because in this learning there is a reward system.

Given a situation…what will be the analysis…

sallypmangaba — Sat, 01 Mar 2008 08:06:52 +0000

Supposedly, a traffic management system uses a neural net to predict the flow of traffic during a specific time or schedule. What would be the best way of training this system that it is possible to extract complex data such as employees and students schedule?

I consider supervised training as the answer, because we can have different inputs which are the employees and students schedule which are from reliable sources. In this training, we can find the least cost and produce a minimum error between the data. The data about the schedule of employees and students would be a great help to predict the traffic flow during a specific period of time. As the hidden layer, the traffic management should conduct surveys and monitoring.

When is a reinforced training method best to implement and when is it not most efficient?

sallypmangaba — Sat, 01 Mar 2008 08:05:04 +0000

I believe that Reinforced training is best to implement in studying, we can try so many strategies just to do good in studies. We can have adjustments and we can control problems and be able to have rewards.It is not most efficient on giving medicines to patients or in conducting operations because it is not proper to have trial and error on these situations because it is a matter of life and death.

Why do networks using competitive learning strategies are describe as “self organizing networks/maps?

sallypmangaba — Sat, 01 Mar 2008 08:03:46 +0000

As I have learned, Competitive learning is described as a self-organizing map because this learning law is using unsupervised training, these are sets of neurons and the weights of the winning neuron are adjusted. There is a competition and the neuron whose weight was closest to the input is updated to be even closer. If there are many inputs, every neuron in the layer that is close to the group of inputs will adjust their weights. Just like SOM (self-organizing maps), competitive networks learn to categorized and distribute the input vectors.

LEARNING LAWS and THE TRAINING RELATED TO IT

sallypmangaba — Sat, 01 Mar 2008 08:02:11 +0000

a. Coincidence

I consider here that the training approach is supervised training because the input and output are known. The weights are adjusted. If there is an input it will find an output to another neuron. There is a linear association and series of inputs. Just like the domino effect, one after another, and then many inputs may produce one output.

b. Performance

In this learning law, I consider the use of supervised training because this continuously modifies the strengths of the input connections to reduce the difference between the desired output value and the actual output of a processing element. This also changes the weights. The error in the output layer is transformed by the derivative of the transfer function and is then used in the previous neural layer to adjust input connection weights.

c. Competitive

This is using unsupervised training. There are sets of neurons, the inputs are unknown but the outputs are known, there is adjusting of the weights to produce the desired winner as the output. Winning elements are allowed to modify their weights thus competition takes place to see which unit has the small input intensity.

d. Filter

This is using reinforced training because of there is a preset filter, all signals will compute the element which is a time-based, meaning it adjusts the signal base on time, the signals that failed will be dropped.