Intelligent Algorithm
"Algorithm" is originally another word for a recipe for solving a mathematical problem. Because the function of the objects that work using learning software are often based on mathematical and statistical models, the concept of algorithm has also penetrated here. To indicate that the algorithm is deployed and functions on the basis of learning software, the concept of intelligent algorithms is also becoming increasingly common. This concerns the algorithms that, for example, are continuously trained and validated and therefore become increasingly "smarter".
Algorithms do not have human cleverness (for example something like intuition), but work on the basis of performing calculations and can be optimized in such a way that it seems as if they are becoming "smarter". A better word for that is accuracy. The percentage of incorrect predictions, for example, is decreasing. We strive to make the algorithm work more accurately.
The added value of an Intelligent Algorithm can be enormous in a financial sense. Making better and faster decisions has major advantages. The costs for setting up and further optimizing algorithms are determined by the investment in relation to the expected yield. It is important to determine in advance whether the function of the Intelligent Algorithm can live up to expectations. We demonstrate this by first developing a prototype and thoroughly testing it in practice.
The Intelligent Algorithm must be able to work in the background. In many cases this involves the execution of a single task in a process. For example, categorizing something (right / wrong) or predicting a value based on collected data. In general, we do this in order to be able to make a better decision. Perhaps to prevent something from going wrong in the process. People can then decide to intervene preventively.
There are therefore many possibilities for which an Intelligent Algorithm can be used. However, we are also dealing with the chicken or egg problem here. If we don't know that an Intelligent Algorithm can help solve a particular problem, we don't need it either. We don't ask the question because we don't know what we don't know. In order to ask these questions, it requires some awareness about the possibilities and applications of Intelligent Algorithms. Zuiver ICT helps to ask the most important questions and helps to make a good choice for a type of Intelligent Algorithm and the steps that need to be taken.
In essence, it is about setting up a software application in a different way than before. A "classic" software program consists of fixed rules, the program, with which you indicate exactly what the software should do and how it should respond to the input provided. The program follows a pattern of predefined rules. An Intelligent Algorithm also works on basic software, but is actually largely programmed by "training" it first.
Our principles for developing, testing and operationalizing an Intelligent Algorithm are:
- The simpler the better
- Relatively easy to embed in the operation
- Robustly designed, minimal technical maintenance
- Non-discriminatory
- Transparent in functioning
Operationalize Intelligent Algorithms
The challenge in the technical sense arises when operationalizing the Intelligent Algorithm. We can consider the Intelligent Algorithm as a "forecasting machine" that is only turned on when it is needed. When it is started, it has to perform its function very quickly and offer the output where we want it. In addition, we also set requirements for the manageability and transparency of the Intelligent Algorithm. To organizations that do not have sufficient options to operationalize an Intelligent Algorithm in their own IT environment, Zuiver ICT provides the option of offering the Intelligent Algorithm function as a Service from the Cloud.
If we see the possibility of adding a lot of value in a simple way, then that is our preference. Complexity is not a goal, "Simplicity is the soul of efficiency".
Something about Deep Learning and Reinforcement Learning
The power of Intelligent Algorithms has increased enormously in recent years through the application of Deep Learning and Reinforcement learning. Deep Learning is a sub-category within Machine Learning and works under the hood on the basis of a neural network. The result is achieved in a similar way to how the human brain works with neurons. Zuiver ICT also tests the application of Deep Learning in the development of an Intelligent Algorithm to perform a specific function. It is not always clear whether a Deep Learning solution is the right solution, mainly due to the increase in complexity.
Reinforcement Learning is an emerging, new way of Machine Learning in which the Intelligent Algorithm is no longer "classically" trained by designing the software based on historical data, but the algorithm learns from the mistakes it makes. This form of "training" is currently receiving more and more attention. Zuiver ICT follows the developments in this respect and will also apply this where possible, provided this method of "training" also offers sufficient added value for the issue we are facing. The point is not that a solution is "state of the art", but that the solution works in practice and is sufficiently robust.