Gradient descent and backpropagation enable adjustments like these in an artificial neural network. If an artificial neural network were learning to expand its menu of jellybeans, it would start out with white jellybeans and would move along the color gradient to sample darker and darker jellybeans. The network would test the flavor of each until it eventually tasted a black jellybean, at which point, the backprop algorithm would kick in and tell the neural network that it had gone too far on the color gradient. (Keep in mind that backprop is typically only used for supervised learning.)
These algorithms work by making adjustments to the weights of each connection. Each neuron in an artificial neural network has a weight (typically zero through one), which shows its strength to a previous or succeeding neuron. The closer the weight is to the number one the stronger the connection. The closer the weight is to zero the weaker the connection. A neural network adjusts the weights of these connections over time as a way to match different patterns. A strong connection shows a clear match. A weaker connection shows only a possible match or no match at all.
With supervised learning, you need a way to let the neural network know when it has made a mistake when it has failed to identify a match or has falsely identified a match. Suppose the neural network mistakes a purple jellybean for a black jellybean. The backprop algorithm tweaks the weight of the neural connections to reduce the possibility that the neural network will make this same mistake in the future.
Remember that my friends and family had to coax me into trying darker color jellybeans. The same is true with an artificial neural network. A human being has to identify the white and black jellybeans and then help the network twist the dials of the gradient to expand its menu of acceptable jellybeans.
Classification isn’t the only form of supervised machine learning digital media strategy. You can also have your artificial neural network use something called regression analysis, the purpose of which is to identify the relationship between a dependent variable and one or more independent variables.
To understand regression analysis, imagine those sausage shaped balloons you see at children’s parties. You squeeze one end, and the other end bulges. If you let it go, the balloon returns to normal. If you squeeze both ends, the center bulges. Release one end, and the bulge moves to the opposite end. Each squeeze is an independent variable. Each bulge is a dependent variable; it differs depending on where the balloon is squeezed.
Now think about what happens at your typical children’s party. Some wacky balloon performer will twist together five or six of these balloons to create various balloon animals. Now the relationship between squeezing and bulging is much more complex. If you squeeze the body, maybe the tail bulges.
If you squeeze the head, maybe two legs bulge. Each change to the independent variable results in a change to one or more dependent variables. Sometimes that relationship is easy to predict, and other times it’s extremely difficult.
I once worked for credit card processing organization that was trying to look for warning signs of when a customer would have trouble paying their bill. They used a regression in their artificial neural network to try to find relationships between different variables. What they found is that many customers start to put essentials on their credit card just before they have trouble paying their bill. A customer who typically used their card only for large purchases, such as a television or computer, would suddenly start using it to buy groceries and gas and pay their electric bill. The credit card company also found that people who had a lot of purchases under five dollars were likely to have trouble paying their bill.
The dependent variable was whether the person would have enough money to pay the credit card bill. The independent variables were the items the customer purchased and the payment amounts. The artificial neural network identified a relationship between the dependent and independent variables that provided valuable insight to the credit card company.
The dependent variable in the case of classification is a “label” or category, such as cat, dog, apple or jazz music. The dependent variable in regression is mostly a numerical value, such as income, height, weight or population.
Even though the information you get from regression analysis differs from the information you get through backpropagation, the way you use your artificial neural network is similar. It still requires massive amounts of data to identify patterns. The network examines this data to identify patterns that a person may never have thought to look for. The difference is that backprop (with neural networks) is a classification algorithm. So you’re using classification to predict a label or category. In regression, you find relationships between independent and dependent variables so the neural network can predict the behavior of a dependent variable.
When you’re starting your own AI project, you need to consider whether the job requires sorting and connecting. If it requires sorting, backpropagation is the better choice. If it requires identifying connections, go with regression analysis.
Keep in mind, however, that whether you’re using backpropagation or regression analysis, your artificial neural network can only show you the patterns. It doesn’t necessarily provide the answers, and it doesn’t offer explanations. For example, in the credit card example, the network pointed out that when customers started using their cards for purchases of five dollars or less, they were likely to have trouble paying their credit card bill, but it didn’t explain why. After the network identified the pattern, it was up to human beings to determine why.
Perhaps at some point these artificial neural networks will be able to create their own theories for why these patterns exist. For now, it’s up to humans to find meaning in these connections.
via WordPress https://managingzen.wordpress.com/2018/11/30/regression-analysis-on-ai/