Neuronal Networks and Machine Learning

Have you already trained a neural network?

Have you got children?

Neural networks learn in the same way as children: Through trying and receiving feedback. The feedback scoring adjusts the behavior to achieve better results the next time you try. In computer science, the "machine learning" describes a class of algorithms that independently modifies its own parameters in order to achieve better solutions. A subset of machine learning are neural networks, which are becoming more and more widespread in current systems. Ideal use cases for neural networks are systems with few degrees of freedom based on certain patterns. Examples such as "3 dimensional strip-packing problem " from logistics and classification and completion in image processing are excellent use cases.

Neural networks learn in the same way as children: Through trying and receiving feedback. This modifies the behavior and leads to better results. See how neural networks can be applied in a purposeful way.

Three-dimensional packaging problem

Items of different dimensions should be packed into a box with postage-optimized sizes. A statistical algorithm solves this task with randomly directed steps.

Neural networks detect packaging patterns and solve the problem within a far shorter time.

Classification and Completion

CNNs, or Convolutional Neural Networks, are suitable for detecting people or classifying items. As a result, you get the detected objects rated by a detection probability.

Items can be removed from images and replenished with detected patterns.

When do we use Neural Networks?

Neural networks detect selected patterns in an unbeatably efficient manner. Therefore, they are the ideal complement to traditional deterministic and statistical algorithms.

Even a simple problem of packing a number of items optimized in boxes demands large computing time from a deterministic algorithm (see "3-dimensional strip-packaging problem").

In order to decrease computing times, it is a good idea to try random solutions and evaluate them against defined criteria. The statistical algorithms working this way can be used to train neural networks.
The result evaluation of the neural network is carried out with the help of the statistical algorithm. With this feedback, the neural network is trained in order to find optimal solutions in the shortest possible time.