|
||
We can calculate the conditional probabilities the same way as we calculated the prior probabilities: By observing the events around us! For example, we may have to search old newspapers and find out all the times somebody fell from the 10th floor while there was a safety net underneath, and count the times he died or survived! Equally, we may do statistics on the fraction of forests that are in Attica and get burned. |
||
| The trouble is that in some cases it is difficult to find enough
events of the right type to do reliable statistics with. We say that "we
need training data". The more training data we have, the better our statistics
are, and the more sound our predictions become.
Systems which can learn from a small number of data are systems with "superlearning capabilities". Humans have such capabilities as they learn from a few examples. Some Neural Networks have such capabilities too. (The trouble is we do not know which have and which do not have!) Stassopoulou and Petrou tried to combine a probabilistic reasoning system with a neural network to exploit the learning capabilities of the latter. However, more of that later. |
||
|
||
|
||
 |
|
|
