News

Machine Learning and Deep Learning - A look at future technologies

What are machine learning and deep learning actually about? And what benefits do these future technologies offer specifically for logistics?

The “From the laboratory of the future” feature presents findings from the Research & Development division, which works in close collaboration with various departments and branches, as well as the Dachser Enterprise Lab at Fraunhofer IML and other research and technology partners.

As with anything that gets hyped, machine learning technology is not about to solve all our problems and therefore will not change everything. That’s why most existing IT systems in logistics will not be replaced by machine learning and hence artificial intelligence. But the technology does have the potential to address problems that conventional programming logic has so far been unable to solve, prime among them image, text, and speech recognition, the interpretation of complex data volumes, and predictive analytics. These are all areas that offer possible applications in logistics: for example, the forecasting of volume and price developments, the classification of packages using images, the interpretation and automatic further processing of unstructured input data (e.g., e-mail inquiries), and the operation of autonomous vehicles and machinery in changing working environments.

Independent learning phases and traditional mathematical techniques

When using typical if-then-else programming in these use cases, it would be necessary to think of all eventualities in advance and convert them into lines of code. However, this is not usually possible due to the complexity and amount of data. Machine learning takes a different approach: an algorithm trains itself automatically based on historical input data. This process, which is also known as the learning phase, can be said to have been a success once the algorithm is able to calculate the desired output data for similar but unknown input data. The algorithm has then independently found a regularity through its training. 

Machine learning refers to a whole range of traditional mathematical techniques, such as decision trees or what are known as k-means (clustering). Another approach involves artificial neural networks (ANNs), which realize a type of abstraction model of the human brain and are inspired by human learning behavior. Many developers currently rely on deep learning—i.e., ANNs with a large number of neuron layers—to deal with high complexities. Especially here, they are trialing various methods and tools.

Machine learning is a promising technology, and its initial applications are encouraging. However, it is still in a development phase. When and to what extent it will change supply chains will become apparent only in the next few years.

Contact Christian Weber Corporate Public Relations