Introduction to Cindicator (CND) Part 3

Deep learning continued.

 

Deep learning can establish correlations between past, present and future events in the same way it can establish correlations between pixels in an image or a video and the identity of a person. For deep learning, there is no difference between predicting future and deriving a result from set of some data because in a way, a future event for deep learning is a label that in needs to create just like an identity of a person can be a label it needs to create. For software and machines, time does not have the same meaning as it does for people. For computers, the fact that something did not yet happen does not have an emotional element to them. Identifying potential outcomes of what may happen in the future to a computer is the same as solving any other kind of problem, be it performing a mathematical computation, identifying a traffic sign in an image or coming up with probabilities and outcomes of future events.

Given a set of data with a timeline, deep learning can predict hardware breakdowns, potential health issues such as heart attacks, the probability that a customer is going to leave, the probability that an employee is going to leave the company or stay with the company, and so on.

 

How deep learning can help businesses

The more clarity organizations can have about what is happening with their systems, processes, customers and employees, the better job they can do to prevent negative events and to reinforce positive events.

For example, according to the data from the United States Small Business Administration, most customers stop patronizing a business because they feel that the business doesn’t really care about them. It may be that they have brought a friend into a store after recommending the brand. The friend made an expensive purchase, but the brand did not link the two names in the database, did not thank the referring customer in any way and the customer is leaving because he or she does not feel valued. It may be that competition starts showing ads to the customer and creates a special offer trying to lure the customer away from the brand. If the brand does not about what is going on, it is likely that the customer is going to leave. If the brand does know what is happening, it can start a series of preventive actions designed to keep the customer.

In the case with the customer bringing in friends to a brand’s store, the company could send a post card with a thank you note and some small gift or coupons towards future purchases. In case with competitors showing the customer ads, the company could launch its own marketing campaign and explain why the customer should stay with the company.

This approach could work in the same way in a number of industries and be extremely beneficial both to organizations and their clients.

 

Deep learning in various industries

For instance, in healthcare deep learning could predict potential health issues and doctors could explain to patients what prevention steps the patients needed to take, be it increase in exercise, change in diet, or introduction of a new medication into their life.

With potential hardware breakdowns, data centers could be replacing hardware before the hardware actually breaks down. They could also be creating excessive backups if deep learning was to predict that a system or a part of a system was about to fail.

In transportation and cargo industries, companies could ground vehicles and vessels before they break down and transfer cargo and passengers from one vehicle or vessel to a different one of redirect a vessel from one port to a different one, the one that is located closer to the current destination of the vessel to increase the probability that the vessel gets there before it breaks down and the cargo stays safe and undamaged.

Such new behavior will benefit all the players and save money for everyone, from the organizations to their customers to insurance companies.

Typically, computation on a deep learning (or neural) network happens in a node. A node on a neutral network is loosely similar to a neuron in human brain and combines information with sets of various weights that can amplify or lessen different criteria present during the decision-making process. After summarizing the weights, the node sums them up and passes them through an activation function that establishes what it should do next with the signal. It can choose to use it in the decision-making process, ignore it, or add it to the existing classification, which is the learning element of the neural networks.