Artificial intelligence, machine learning, data mining. Laura, as an expert, can you put these terms into some sort of order?
Essentially, all these terms describe the same objective – namely to extract knowledge from data and to make it usable for us humans. The term data mining is used when the aim is to extract patterns from large quantities of data using statistical methods and to identify connections between them. We talk about machine learning when intelligent algorithms are being used which automatically detect such patterns and can use this knowledge independently to solve tasks. To use such algorithms successfully, it is often necessary, however, to analyse the data in advance using statistical methods and to prepare it for the algorithm.
We only refer to artificial intelligence, however, when algorithms also handle tasks which previously could only be solved by human capabilities – such as seeing, speaking or learning from experience. These algorithms often come from the subject area of machine learning which is therefore a key sub-area of artificial intelligence.
Machine learning sounds like it is fully automatic. How much human intervention is involved?
Learning algorithms are not a miracle cure which can detect all patterns and connections in our data and provide these to us completely off their own bat. Their potential can only really be unfurled through the interaction of people and machines: we humans provide data and must ensure that it is of a high quality and is suitable for solving a specific task. To this end, we select suitable algorithms, prepare all the features in the data accordingly and organise the entire journey from the source of the data right through to the result of the analysis. If machine learning is to be used successfully, both human creativity and critical human thinking are required.
For our magazine Quelltext, you have worked on a machine learning showcase using e-commerce data. To what extent are your results also relevant for other industry sectors and business cases?
Most of the sections that the showcase describes are generally of significance for machine learning projects, in other words, they can be applied to many application scenarios. Just like artificial neural networks, methods which use gradient boosting can be used very flexibly and are extremely performant. The algorithm we use, Catboost, is not just suitable for forecasting sales figures or expected profits. It can also be used to detect, for example, whether customers are responding to marketing strategies as required or whether and when production machines are malfunctioning. The showcase also provides many points of reference for further application scenarios. The extreme outliers in the sales figures and prices of products make it clear that algorithms for anomaly detection, for example, could help to detect erroneously high orders. Furthermore, customer feedback could help to reduce cancellations and outlay.
In addition, clustering procedures and natural language processing can help us find out which products are similar and concurrently on-trend or are enjoying multiple sales. Customers’ buying behaviour can also be scrutinised for the purposes of customer segmentation.
Which particular application scenarios of machine learning do you think could result in progress for society in general?
Particularly in the area of medicine, there are very many useful application scenarios which are conceivable. Deep learning can not only be used to detect types of cancer at an early stage in medical image data, for example. I can also see artificial neural networks helping here to assist with ECG monitoring. Is someone about to suffer a heart attack? The signals may already reveal minor changes which learning algorithms can detect although the human eye cannot perceive them.
Dear Laura, thank you very much for the interview!
Are you interested in the topic of machine learning? Then you should also read Laura Fink’s related blog or the sample application, VAMINAP. Here in addition is an interview with Laura Fink in the special supplement “Artificial Intelligence” of the German Handelsblatt, page 7.